feat(teams): add Strategic Ops, Cost Ops teams and agent analysis engine#356
Merged
JustAGhosT merged 16 commits intodevfrom Mar 11, 2026
Merged
feat(teams): add Strategic Ops, Cost Ops teams and agent analysis engine#356JustAGhosT merged 16 commits intodevfrom
JustAGhosT merged 16 commits intodevfrom
Conversation
- BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…mplate hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment
Contributor
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
WalkthroughAdds a new agent-analysis engine and tests, an analyze-agents CLI command with dynamic flag loading and wrappers, new teams/agents in spec files, sync integration to emit agent-team matrices, many generated scripts/docs/personas/rules, and assorted script/CI tweaks and test adjustments. Changes
Sequence Diagram(s)sequenceDiagram
participant User
participant CLI as CLI (.agentkit/engines/node/src/cli.mjs)
participant Spec as SpecLoader
participant Analysis as AgentAnalysis
participant Renderer as Renderer
participant FS as FileSystem
User->>CLI: run "analyze-agents" (--matrix, --format, --output)
CLI->>CLI: loadCommandFlags(AGENTKIT_ROOT)
CLI->>Spec: read `spec/commands.yaml`, `spec/agents.yaml`, `spec/teams.yaml`
Spec-->>Analysis: parsed specs & command flags
Analysis->>Analysis: build graphs (depends/notifies/handoffs)
Analysis->>Renderer: compute matrices / analyses
Renderer-->>Analysis: formatted Markdown or JSON
Renderer->>FS: write output (file or stdout)
FS-->>User: deliver output
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related PRs
Poem
✨ Finishing Touches🧪 Generate unit tests (beta)
|
AgentKit Forge Source Change DetectedThis PR modifies files in the AgentKit Forge source directories:
These are the upstream source-of-truth for all generated AI tool configs. Review checklist
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 17
Note
Due to the large number of review comments, Critical, Major severity comments were prioritized as inline comments.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (4)
.github/workflows/claude.yml (1)
15-20:⚠️ Potential issue | 🔴 CriticalDo not use self-hosted runners for comment-triggered workflows.
This job triggers on comments and reviews in public contexts (
issue_comment,pull_request_review_comment). GitHub's security guidance explicitly states: "Self-hosted runners should almost never be used for public repositories" because anyone can open PRs or post comments to compromise the runner environment. Even if secrets aren't automatically passed to fork-based PRs, the runner itself and the host infrastructure become exploitable. Either use GitHub-hosted runners or implement strict gatekeeping: runner groups with access restrictions, trusted actor checks, and a dedicated isolated runner pool.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.github/workflows/claude.yml around lines 15 - 20, The workflow uses a self-hosted runner for comment-triggered events (see the if condition and the runs-on: self-hosted in the claude.yml job), which is insecure for public comment triggers; change runs-on to a GitHub-hosted runner (e.g., ubuntu-latest) or, if you must use self-hosted, restrict execution by adding strict gatekeeping such as verifying the actor is trusted (check github.actor in the if expression), limiting to a dedicated isolated runner group with limited permissions, and disallowing forks/comments from untrusted actors before allowing the job to run.scripts/setup-agentkit-branch-governance.sh (1)
106-114:⚠️ Potential issue | 🟡 MinorBug: Branch protection applied twice to
main.The loop iterates over
main main, applying identical protection rules to the same branch twice. This appears to be a find/replace artifact wheredevwas replaced withmainin what was originallydev main.If the intent is to protect only
main, remove the duplicate. If bothmainand another branch (e.g.,devordevelop) need protection, update accordingly.🐛 Proposed fix
if [[ "$SKIP_PROTECTION" == false ]]; then - for BRANCH in main main; do + for BRANCH in main; do if [[ "$DRY_RUN" == true ]]; then echo "[dry-run] Would apply branch protection to $REPO/$BRANCH" continue🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@scripts/setup-agentkit-branch-governance.sh` around lines 106 - 114, The for-loop currently iterates "for BRANCH in main main" which applies branch protection twice; update the loop to list the intended branches (e.g., "main" only or "main dev"/"main develop") so each branch is processed once. Locate the loop that sets BRANCH and adjust the second token (remove the duplicate or replace with the correct branch name), leaving the existing DRY_RUN check and the gh api --method PUT "/repos/$REPO/branches/$BRANCH/protection" call intact..agentkit/engines/node/src/__tests__/cli.test.mjs (1)
130-156:⚠️ Potential issue | 🔴 CriticalFile is in a protected .agentkit/ directory and requires CODEOWNERS approval.
While
analyze-agentsis fully implemented in the CLI (defined in commands.yaml, has flags and handler logic at cli.mjs:53, :112, :392, :661), it is missing from the test commands array at lines 130-156. However,.agentkit/engines/is an upstream source-of-truth protected directory. Any modifications to this file require CODEOWNERS approval per project governance.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/engines/node/src/__tests__/cli.test.mjs around lines 130 - 156, The test is missing the 'analyze-agents' command in the commands array used by the CLI tests; update the commands array (the variable named commands in the test) to include 'analyze-agents' (position it to mirror commands.yaml/cli.mjs ordering), run the test suite to confirm it passes, and because this file lives in a protected agentkit directory, include the change in your PR and request CODEOWNERS approval before merging..agentkit/spec/commands.yaml (1)
711-738:⚠️ Potential issue | 🟠 MajorMake
--generate-planstruly opt-in.Line 711 says Phase 2.5 runs only when the flag is passed, but Lines 735-738 default the flag to
true. With a positive boolean flag and no--no-generate-planscounterpart, omitting the flag still scaffolds plans and turnsproject-reviewinto a mutating command by default.Minimal fix
- name: --generate-plans description: 'After Phase 2, scaffold plan files from critical/high findings into docs/planning/review-findings/' type: boolean - default: true + default: false🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/spec/commands.yaml around lines 711 - 738, The --generate-plans flag is currently defaulting to true, making plan scaffolding run unless explicitly disabled; change the flag definition for --generate-plans in commands.yaml so its default is false (and keep required: false) and update the flag description to state it is opt-in (e.g., "opt-in: scaffold plan files only when provided"); this ensures that command behavior is non-mutating by default and that scaffolding only occurs when --generate-plans is explicitly passed.
🟡 Minor comments (9)
scripts/resolve-merge.ps1-1-7 (1)
1-7:⚠️ Potential issue | 🟡 MinorFix the UTF-8 BOM issue in the upstream template.
This generated PowerShell script contains non-ASCII em dashes (lines 1, 5, 45, 52, 141, 144) without a UTF-8 BOM, which causes encoding compatibility issues in Windows PowerShell 5.1. Since this file is marked
GENERATED by AgentKit Forge — DO NOT EDIT, modify the template in.agentkit/spec/to emit either a UTF-8 BOM or use ASCII-only em dash alternatives in comment headers, then runpnpm -C .agentkit agentkit:syncto regenerate.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@scripts/resolve-merge.ps1` around lines 1 - 7, The generated resolve-merge.ps1 contains non-ASCII em dash characters causing PowerShell 5.1 encoding issues; fix the AgentKit Forge upstream template that produces this file by either (a) changing the template to emit a UTF‑8 BOM on generation, or (b) replacing the em dashes in the template comment headers with ASCII hyphens or plain text equivalents, then re-run the AgentKit sync/regeneration step so resolve-merge.ps1 is regenerated without the non-ASCII characters..github/workflows/issue-label-validation.yml-101-107 (1)
101-107:⚠️ Potential issue | 🟡 MinorClear automation labels before this early return.
On
editedevents, a previously templated issue can lose its### Area/### Priorityheadings. Returning on Line 104 skips the stale-label cleanup below, soarea:*,priority:*,severity:*, andinvalid-fieldscan remain attached even though the issue is now intentionally exempt from validation. Please move the existing cleanup block above this guard, or run it in the no-form branch as well.Also applies to: 152-168
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.github/workflows/issue-label-validation.yml around lines 101 - 107, The early return guarded by the hasFormFields check (const hasFormFields = /### (Area|Priority)/i.test(body); if (!hasFormFields) { core.info(...); return; }) skips label cleanup for edited events; move or invoke the existing cleanup block that removes area:*, priority:*, severity:*, and invalid-fields labels to run before this guard (or call it in the no-form branch) so that stale labels are always cleared when hasFormFields is false; apply the same change to the duplicate block around lines 152-168 so both occurrences call the cleanup routine prior to returning.scripts/setup-agentkit-branch-governance.sh-117-120 (1)
117-120:⚠️ Potential issue | 🟡 MinorDuplicate verification commands.
Lines 119-120 both output the same verification command for
main. If onlymainis protected, keep one line. If multiple branches should be listed, update accordingly.🐛 Proposed fix
echo echo "Done." echo "Verify with: gh api /repos/$REPO/branches/main/protection" -echo " gh api /repos/$REPO/branches/main/protection"🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@scripts/setup-agentkit-branch-governance.sh` around lines 117 - 120, The script prints the same verification command twice for the main branch (the duplicate echo of "gh api /repos/$REPO/branches/main/protection"); remove the repeated line or replace the second echo with the intended branch(s) to verify (e.g., change to another branch name or loop over a BRANCHES variable) so the output only shows the correct set of verification commands for protected branches..agents/skills/analyze-agents/SKILL.md-37-42 (1)
37-42:⚠️ Potential issue | 🟡 MinorMinor: Inconsistent list nesting in Project Context.
Line 41's
Tech stackappears nested underDefault branchdue to indentation, but semantically they should be sibling items. This may be a template generation issue.📝 Suggested structure (modify in source spec)
## Project Context - Repository: agentkit-forge - Default branch: main - Tech stack: javascript, yaml, markdownSince this is a generated file, the fix should be applied in the source spec at
.agentkit/spec/and regenerated viapnpm -C .agentkit agentkit:sync.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agents/skills/analyze-agents/SKILL.md around lines 37 - 42, The "Project Context" section has incorrect list nesting where the "Tech stack" item is indented under "Default branch" (look for the "Project Context" heading and the list items "Default branch" and "Tech stack"); fix the source spec that produces this generated file so "Tech stack" is a sibling list item (remove the extra indentation in the template/spec for that list entry), then re-run the agentkit generation/sync step to regenerate the markdown output.docs/planning/README.md-331-337 (1)
331-337:⚠️ Potential issue | 🟡 MinorRoute stale-marker cleanup through the source specs/templates.
These entries point at
.agentkit/docs/...generated outputs. Telling contributors to remove or update those files directly will just reintroduce drift on the next sync and can fail CI.As per coding guidelines,
Never edit files marked 'GENERATED by AgentKit Forge — DO NOT EDIT'. Modify the source spec in '.agentkit/spec/' and run 'pnpm -C .agentkit agentkit:sync' instead.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@docs/planning/README.md` around lines 331 - 337, The listed table entries in docs (e.g., .agentkit/docs/getting-started/QUICK_START.md, COMMAND_REFERENCE.md, AGENTS_REFERENCE.md, AGENTS_VS_TEAMS.md, WORKFLOWS.md) are generated and should not be edited directly; instead update the corresponding source spec/template under .agentkit/spec/ (modify the spec that generates the affected section or remove the outdated "In-Flight" entries) and then re-generate the docs with pnpm -C .agentkit agentkit:sync so the changes persist and CI won't be broken.scripts/sync-split-pr.ps1-42-48 (1)
42-48:⚠️ Potential issue | 🟡 MinorEmit real UTC timestamps in the usage log; fix the source spec file.
This file is marked
GENERATED by AgentKit Forge — DO NOT EDIT. The timestamp bug exists in the source; modify.agentkit/spec/and regenerate usingpnpm -C .agentkit agentkit:sync.
Get-Date -Format 'yyyy-MM-ddTHH:mm:ssZ'formats local time and appends a literalZ, so these entries incorrectly claim UTC even on non-UTC hosts, skewing ordering and audit data intool-usage.jsonl.Suggested source fix
Replace all three occurrences (in the source spec) with:
- timestamp = (Get-Date -Format 'yyyy-MM-ddTHH:mm:ssZ') + timestamp = (Get-Date).ToUniversalTime().ToString('yyyy-MM-ddTHH:mm:ssK')Using the
Kformat specifier properly rendersZfor UTC timestamps.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@scripts/sync-split-pr.ps1` around lines 42 - 48, The generated script sync-split-pr.ps1 emits a local timestamp with a literal "Z"; fix the source spec files under .agentkit/spec by replacing the Get-Date format that appends a literal Z with a format that uses the .NET "K" time zone specifier so timestamps truly represent UTC/offsets, update all three occurrences in the spec, then regenerate the script using pnpm -C .agentkit agentkit:sync to produce corrected sync-split-pr.ps1.scripts/update-changelog.ps1-1-3 (1)
1-3:⚠️ Potential issue | 🟡 MinorUse ASCII in generated PowerShell banners or emit UTF-8 BOM.
The new banner adds a non-ASCII em dash on lines 1-3, and the file is UTF-8 without BOM. This combination is incompatible with Windows PowerShell 5.1. Both
scripts/update-changelog.ps1andscripts/analyze-agents.ps1exhibit this issue. Either switch the banner text to plain ASCII (-) or configure the generator to emit UTF-8 BOM for generated.ps1files.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@scripts/update-changelog.ps1` around lines 1 - 3, The UTF-8 file header uses a non-ASCII em dash which breaks PowerShell 5.1; fix by either replacing the em dash with a plain ASCII hyphen in the generated banner comment lines of scripts/update-changelog.ps1 and scripts/analyze-agents.ps1, or configure the AgentKit Forge generator to emit a UTF-8 BOM for generated .ps1 files (update the generator config that produces the banner in .agentkit/spec or .agentkit/overlays/agentkit-forge) so PowerShell can read the non-ASCII character..windsurf/rules/team-strategic-ops.md-8-8 (1)
8-8:⚠️ Potential issue | 🟡 MinorRemove backslash escape from glob pattern.
The pattern
**/\*is inconsistent with all other glob patterns in the codebase, which use**/*. The backslash escape (\*) is markdown syntax that prevents the asterisk from being interpreted as markdown formatting, but it will break the actual glob pattern matching in Windsurf Cascade's rule processor. Change to**/*or remove the markdown escaping.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.windsurf/rules/team-strategic-ops.md at line 8, The glob pattern contains a markdown-escaped asterisk (`**/\*`) which will break Windsurf Cascade matching; update the pattern string to use an unescaped asterisk (`**/*`) so it matches directories/files correctly (look for the `**/\*` token in the rule and replace it with `**/*`)..agentkit/templates/scripts/update-changelog.sh-65-67 (1)
65-67:⚠️ Potential issue | 🟡 MinorPR link uses incorrect relative path.
The link
../../pull/${PR_NUMBER}is relative to the CHANGELOG.md file location. If CHANGELOG.md is at repo root, this path would resolve incorrectly. GitHub PR links should use an absolute URL or a root-relative path.🔧 Proposed fix
if [[ -n "$PR_NUMBER" && -n "$HISTORY_DOC" ]]; then - ENTRY="$ENTRY ([#${PR_NUMBER}](../../pull/${PR_NUMBER}), [history](${HISTORY_DOC}))" + ENTRY="$ENTRY ([#${PR_NUMBER}](https://github.com/\${GITHUB_REPOSITORY:-OWNER/REPO}/pull/${PR_NUMBER}), [history](${HISTORY_DOC}))" elif [[ -n "$PR_NUMBER" ]]; then - ENTRY="$ENTRY ([#${PR_NUMBER}](../../pull/${PR_NUMBER}))" + ENTRY="$ENTRY ([#${PR_NUMBER}](https://github.com/\${GITHUB_REPOSITORY:-OWNER/REPO}/pull/${PR_NUMBER}))"Alternatively, if you want to keep relative links, consider using a format that works from the repo root like
./pull/${PR_NUMBER}or fetching the repo URL dynamically.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/scripts/update-changelog.sh around lines 65 - 67, The PR link in update-changelog.sh builds ENTRY using a relative path ("../../pull/${PR_NUMBER}") which will break if CHANGELOG.md is at repo root; update the string construction that uses ENTRY and PR_NUMBER (and HISTORY_DOC) to use either a root-relative path like "/pull/${PR_NUMBER}" or a full absolute GitHub URL (e.g., "https://github.com/<owner>/<repo>/pull/${PR_NUMBER}") or dynamically derive the repo URL, so the generated link resolves correctly from the changelog location.
🧹 Nitpick comments (6)
.agentkit/templates/claude/hooks/protect-templates.ps1 (1)
38-40: Consider logging suppressed parse errors.The empty catch block is flagged by PSScriptAnalyzer. While the intent to proceed with protection on parse errors is reasonable, silently swallowing exceptions can hinder debugging.
📝 Optional: Add debug logging
} catch { - # Ignore parse errors, proceed with protection + # Ignore parse errors, proceed with protection + Write-Verbose "Could not parse package.json: $_" }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/claude/hooks/protect-templates.ps1 around lines 38 - 40, The empty catch block that swallows parse errors should log the exception details instead of being silent; inside the catch following the template parse try (in protect-templates.ps1 / the catch block associated with the template parsing logic), emit a debug/verbose log including the caught error ($_ or $_.Exception.Message and optionally $_.Exception.StackTrace) via Write-Verbose or Write-Debug (or Write-Error at Verbose level) so parsing failures are recorded while still allowing protection to proceed..agentkit/templates/scripts/sync-split-pr.sh (1)
98-110: Add error handling for git push and PR creation.Lines 101 and 109 can fail (auth issues, network problems, branch exists) but errors aren't captured with meaningful messages. Consider wrapping in conditionals or adding
|| { echo "..."; exit 1; }guards.Proposed fix with error handling
-git push -u origin "$NEW_BRANCH" +if ! git push -u origin "$NEW_BRANCH"; then + echo "Failed to push branch $NEW_BRANCH" >&2 + exit 1 +fi -PR_URL="$(gh pr create --base "$BASE_BRANCH" --head "$NEW_BRANCH" --title "$PR_TITLE" --body "$PR_BODY")" +PR_URL="$(gh pr create --base "$BASE_BRANCH" --head "$NEW_BRANCH" --title "$PR_TITLE" --body "$PR_BODY")" || { + echo "Failed to create PR" >&2 + exit 1 +}🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/scripts/sync-split-pr.sh around lines 98 - 110, The script currently runs git push and gh pr create without checking failures; add explicit error handling after the git push of "$NEW_BRANCH" (the git push -u origin "$NEW_BRANCH" command) and after PR creation (the PR_URL assignment using gh pr create) by testing their exit status and printing descriptive errors including context (branch name, remote, and hint about auth/network) and exiting non‑zero on failure; ensure you reference NEW_BRANCH/COMMIT_MESSAGE in the error output for the commit/push step and PR_BODY/PR_URL for the PR creation step, e.g., wrap the git push and gh pr create calls with conditionals or use "|| { echo '...'; exit 1; }" to fail fast with meaningful messages..agentkit/templates/scripts/update-changelog.ps1 (1)
66-106: Embedded Node.js approach is pragmatic but consider extracting.The embedded Node.js script for multiline editing is functional, but embedding ~40 lines of JS in a here-string reduces maintainability. Consider extracting to a standalone
.mjsfile in the templates that can be reused by both.shand.ps1variants.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/scripts/update-changelog.ps1 around lines 66 - 106, Move the embedded JS held in the $NodeScript here-string into a standalone template module (e.g., templates/scripts/update-changelog.mjs) and update the .ps1 and .sh callers to run node against that file instead of embedding the script; specifically, extract the code block that defines changelogPath/section/entry, file I/O and splice logic into update-changelog.mjs, then replace the $NodeScript here-string in update-changelog.ps1 with logic that calls node update-changelog.mjs <changelogPath> <section> <entry> (and make the same change in the shell variant), ensuring the new module exports no state and preserves the same CLI argument behavior and output messages so functions like the existing file-read/write and console.log/console.error behavior remain identical..agentkit/engines/node/src/__tests__/discover.test.mjs (1)
59-59: Rename this spec to the repo'sshould ...style.The new title is clear, but it doesn't follow the test naming convention used for this repo. Based on learnings, "it/test blocks must describe the expected behaviour in plain English using the format
should <verb> <expected outcome>."🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/engines/node/src/__tests__/discover.test.mjs at line 59, The test description for the spec in the it('returns a discovery report with expected structure', async () => { ... }) does not follow the repo convention; rename the string to start with "should" and describe expected behaviour in plain English (e.g. change to "should return a discovery report with the expected structure") by updating the it(...) call in the discover.test.mjs file so the spec reads like "should <verb> <expected outcome>" while keeping the async test body and assertions unchanged.docs/agents/agent-team-matrix.md (1)
30-48: Consider documenting unassigned agents.Matrix 1 shows 16 agents without team mappings (engineering, design, marketing, operations, project-management, feature-management categories). While this may be expected for this PR's scope, consider adding a note in the planning docs about whether these agents will be assigned to teams in future work or operate independently.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@docs/agents/agent-team-matrix.md` around lines 30 - 48, The docs list many agents with no team mapping (e.g., backend, frontend, data, devops, infra, brand-guardian, ui-designer, content-strategist, growth-analyst, dependency-watcher, environment-manager, security-auditor, retrospective-analyst, project-shipper, release-manager, feature-ops); add a short note in the planning docs (or in this agents/agent-team-matrix.md) stating whether these unassigned agents are intentionally teamless, planned to be assigned in future work, or will operate independently, and if applicable add a TODO or link to the roadmap/issue tracking the future assignment for each category..agentkit/templates/scripts/update-changelog.sh (1)
76-77: Consider sanitizing inputs passed to the Node.js heredoc.The
$ENTRYvariable is passed directly to Node.js via command-line arguments. Whileprocess.argvhandles this safely, the entry content could contain characters that might cause issues in the changelog file itself (e.g., unescaped markdown).💡 Optional: Add basic validation
+# Basic validation - reject entries with problematic characters +if [[ "$DESCRIPTION" =~ [\`\<\>] ]]; then + echo "Warning: Description contains characters that may need escaping in markdown" +fi + node - "$CHANGELOG" "$SECTION" "$ENTRY" << 'NODEEOF'🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/scripts/update-changelog.sh around lines 76 - 77, Sanitize and validate the inputs extracted from process.argv (const [,, changelogPath, section, entry] = process.argv) before using them to update the changelog: add checks to trim and reject empty entries, strip or escape problematic control characters and unescaped markdown sequences (e.g., backticks, leading hashes, HTML tags) in entry, and validate section against an allowed set; perform similar minimal validation on changelogPath to avoid path traversal and ensure it points to the expected file type. Implement these checks at the top of the Node heredoc (where process.argv is parsed) and fail early with a clear error when validation fails.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In @.agentkit/engines/node/src/agent-analysis.mjs:
- Around line 454-476: findConsolidationOpps currently treats scope overlap as
literal string equality (scopeA vs scopeB), missing overlaps between glob
patterns; change the overlap detection to use a glob matcher (e.g., minimatch or
micromatch) and consider two patterns overlapping if one pattern would match the
other pattern string or vice‑versa (e.g., minimatch.makeRe(p).test(q) ||
minimatch.makeRe(q).test(p)) so cases like "**/*" vs "**/*.test.*" or "docs/**"
vs "docs/planning/**" are detected; update the computation that builds
overlappingScopes and overlapRatio accordingly (use the same matcher when
iterating patterns in scopeA and scopeB inside findConsolidationOpps) and adjust
the companion test expectations to reflect the glob-aware behavior.
In @.agentkit/engines/node/src/cli.mjs:
- Around line 8-9: Move the top-level import and early call that uses js-yaml
into main() so js-yaml is only required after ensureDependencies() completes:
remove the module-scope "import yaml from 'js-yaml'" and the module
initialization call to loadCommandFlags(), and instead import/require js-yaml
and call loadCommandFlags() inside main() after ensureDependencies() succeeds.
Also update the analyze-agents handler so the parsed matrix flag (matrixArg) is
forwarded to both renderAllAsJson and renderAllAsMarkdown (i.e., call
renderAllAsJson(graph, matrixArg) when format is json and
renderAllAsMarkdown(graph, matrixArg) when format is markdown) to ensure the
--matrix flag is respected. Ensure references to loadCommandFlags,
ensureDependencies, main, analyze-agents handler, matrixArg, renderAllAsJson,
and renderAllAsMarkdown are used to find and modify the code.
In @.agentkit/spec/teams.yaml:
- Around line 86-101: The teams.yaml was changed to add the forge and
strategic-ops teams but you didn’t regenerate the derived docs; run the
generator (pnpm -C .agentkit agentkit:sync) to update the generated outputs
(notably .agentkit/templates/root/UNIFIED_AGENT_TEAMS.md) so the new "forge" and
"strategic-ops" entries appear, then commit the regenerated files alongside your
spec change; ensure the team IDs "forge" and "strategic-ops" and their agent
lists are reflected in the regenerated document before pushing.
In @.agentkit/templates/scripts/resolve-merge.ps1:
- Around line 44-50: The script currently treats any non-zero result from git
merge "origin/${Target}" --no-edit (checked via $LASTEXITCODE) as a conflict;
instead, after a failed merge (non-zero $LASTEXITCODE) check for actual unmerged
paths (e.g., via git ls-files -u or git diff --name-only --diff-filter=U) and
only proceed to the "Merge conflicts detected. Applying resolution matrix..."
path if unmerged files exist; otherwise emit an error/Write-Error and exit
non-zero to surface other merge failures (dirty tree, unrelated errors) rather
than falling through to the auto-commit branch that reports a successful merge.
In @.agentkit/templates/scripts/resolve-merge.sh:
- Around line 79-88: The glob matching is broken because the script strips '*'
and uses grep to match patterns; instead iterate conflicted filenames from git
diff --name-only --diff-filter=U and test each against the glob pattern using
the shell's globbing (e.g. replace the pipeline that ends with grep ... with a
loop that reads each $file and uses [[ "$file" == $pattern ]] to decide
matches), keeping the existing logic inside the while loop (git checkout
--theirs, git add, ok, auto_resolved) and leaving GENERATED_PATTERNS and pattern
variable names unchanged.
In @.agentkit/templates/scripts/setup-agentkit-branch-governance.sh:
- Around line 106-115: The script currently attempts to PUT branch protection
for both {{defaultBranch}} and the literal "main", which causes a 404 and exits
under set -e if "main" doesn't exist; modify the loop around BRANCH to first
check that the branch exists (for example call gh api --method GET
"/repos/$REPO/branches/$BRANCH" and skip/continue when that GET returns 404 or
non-200) before performing the PUT, while preserving the DRY_RUN and
SKIP_PROTECTION checks so missing branches are skipped rather than causing the
script to hard-fail.
In @.agentkit/templates/scripts/sync-split-pr.ps1:
- Around line 70-83: Preflight for the GitHub CLI before creating/ pushing the
branch: before running git checkout -b $Branch and git push -u origin $Branch,
verify gh is installed and authenticated (e.g., check command -v gh and run gh
auth status) and exit with a clear error if missing/unauthenticated; only
proceed to git checkout, git push and gh pr create --base $Base --head $Branch
--title $PrTitle --body $prBody after the preflight passes so you never leave a
half-created remote branch if gh is unavailable.
In @.agentkit/templates/scripts/update-changelog.ps1:
- Around line 1-4: The YAML frontmatter (the leading '---' block) in the
PowerShell template update-changelog.ps1 is invalid PowerShell and prevents
execution; do not edit the template file itself—instead escalate to update the
agentkit spec in .agentkit/spec to decide whether the sync pipeline should strip
YAML frontmatter before deploying scripts or the spec should convert frontmatter
into PowerShell comment blocks (<# ... #>) when generating templates; update the
spec and pipeline logic so all templates in .agentkit/templates/scripts/ are
emitted as valid PowerShell (either by removing frontmatter at sync-time or by
migrating frontmatter into PowerShell comments).
In @.claude/commands/team-strategic-ops.md:
- Around line 2-3: The allowlist in .claude/commands/team-strategic-ops.md does
not permit the mkdir/file update operations used by the lock-and-update protocol
(used around lines 93–103, e.g., the mkdir call at line 96), so update the
allowed-tools entry to include the required shell utilities (for example add
Bash(mkdir *), Bash(rm *), Bash(touch *), or broaden to Bash(*) so the workflow
can create and modify files for the lock/state updates) and then re-run agentkit
sync to regenerate any GENERATED headers as needed.
- Around line 58-60: The doc/logic uses non-canonical statuses ("blocked",
"needs-review") in the handoff handling section—replace those with canonical
statuses used elsewhere: when the handoff exceeds MAX_HANDOFF_CHAIN_DEPTH (check
task.handoffHistory.length + task.handoffTo.length against
MAX_HANDOFF_CHAIN_DEPTH) set task.status = "input-required", append the failure
reason to task.handoffContext, and emit the events.log entry; do not introduce
"blocked" or "needs-review". Also ensure examples/reference to task files point
to the canonical lifecycle (use .claude/state/tasks/ and statuses submitted →
accepted → working → completed/failed/rejected) so all handoff code and docs
(mentions of MAX_HANDOFF_CHAIN_DEPTH, handoffHistory, handoffTo, status,
handoffContext, events.log) remain consistent with the runtime protocol.
- Around line 23-27: The catch-all pattern "**/*" in the team scope must be
removed and replaced with a narrow, explicit set of allowed globs and explicit
excludes for protected upstream dirs; edit the line containing
"docs/planning/**, docs/architecture/**, .agentkit/spec/**, **/*" to remove
"**/*", list only the intended repo subtrees this team owns (e.g.,
docs/planning/**, docs/architecture/**, .agentkit/spec/** or other specific
folders you actually own), and add explicit negative patterns to prevent
upstream edits such as ".agentkit/templates/**, .agentkit/engines/**,
.agentkit/overlays/**, .agentkit/bin/**"; after updating the patterns regenerate
the file output so the new scope replaces the old one and ensure the "stay
within your scope" rule is enforceable.
In @.claude/hooks/protect-templates.ps1:
- Around line 29-44: The source-repo detection path in the hook is wrong: fix
the $agentkitPkg assignment so it points to the hidden .agentkit directory by
changing the string '..\..\..agentkit\package.json' to
'..\..\..\.agentkit\package.json' (ensure the backslash and dot are present);
keep the surrounding logic that uses $scriptDir, Test-Path, Get-Content, and
ConvertFrom-Json intact so the agentkit-forge repo bypass works as intended.
In @.github/chatmodes/team-strategic-ops.chatmode.md:
- Around line 17-18: The STRATEGIC OPS chatmode currently claims the repo-wide
glob "**/*" in the scope (seen in
.github/chatmodes/team-strategic-ops.chatmode.md), which improperly makes this
team own the entire repository; remove or narrow that glob in the upstream team
spec under .agentkit/spec (update the STRATEGIC OPS team entry to use explicit
globs like docs/planning/**, docs/architecture/** or a justified, limited set)
or add a written justification in the source spec if the wide scope is
intentional, then regenerate the generated file by running pnpm -C .agentkit
agentkit:sync rather than editing the GENERATED file directly.
In `@CHANGELOG.md`:
- Around line 27-48: The changelog is missing user-facing features from this PR:
add entries under the "Added" section documenting the new analyze-agents
command/analysis flow (reference the CLI command name analyze-agents and its
analysis pipeline), the new team and agent definitions (mention the new team and
agent config files or classes), and the new sync/script scaffolding (reference
the sync scripts and scaffolding modules introduced); update the upstream
changelog source to include concise bullet points for each feature, follow the
repo's changelog format, and re-sync the updated CHANGELOG.md so the release
notes include these items before merging.
In `@docs/planning/intake/intake-agent-proposal.md`:
- Around line 47-54: The table hard-codes paths like `docs/design/**` and
`docs/processes/**` so the intake agent will miss canonical locations; update
the Location column entries (e.g., rows for `ui-spec`, `process`, etc.) to use
the repo's approved taxonomy under `docs/` (product, architecture,
orchestration, api, operations, engineering, integrations, reference) and ensure
the intake-agent config that consumes these table values watches those canonical
folders instead of `docs/design/**` or `docs/processes/**`; specifically replace
`docs/design/UI-*.md` with the appropriate `docs/product` or `docs/architecture`
path for UI specs, replace `docs/processes/FLOW-*.md` with `docs/orchestration`
(or `docs/operations` if applicable), and audit all table rows to map each Short
ID (`ui-spec`, `process`, `tech-spec`, etc.) to the correct canonical `docs/`
location so the agent indexing logic (the code that reads this table) will pick
up documents stored in the approved structure.
In `@scripts/resolve-merge.ps1`:
- Around line 7-15: This file is generated by AgentKit Forge and should not be
edited directly; instead update the source spec in .agentkit/spec/ and any
overlays under .agentkit/overlays/agentkit-forge and then run pnpm -C .agentkit
agentkit:sync to regenerate the scripts; also update the default merge target by
changing the Param default for $Target (currently "main") to "dev" in the
generator/source spec so the generated scripts use dev as the default target.
In `@scripts/setup-agentkit-branch-governance.ps1`:
- Around line 41-44: The script hard-codes 'main' and duplicates it in the
protection loop; replace the hard-coded 'main' with a parameter/variable (e.g.,
$DefaultBranch or $ProtectedBranch) and use that variable both when PATCHing the
repo default branch (the gh api PATCH call currently using -f
default_branch='main') and when building the protection list (remove the
duplicate entry currently in @('main','main') so the loop only applies each
branch once); update the variable usage where the default is set and in the
protection loop (referencing $Repo, the gh api PATCH call, and the array used
for branch protection around lines 41-44 and 75-81) so overlays can specify a
different default branch and duplicate calls are eliminated.
---
Outside diff comments:
In @.agentkit/engines/node/src/__tests__/cli.test.mjs:
- Around line 130-156: The test is missing the 'analyze-agents' command in the
commands array used by the CLI tests; update the commands array (the variable
named commands in the test) to include 'analyze-agents' (position it to mirror
commands.yaml/cli.mjs ordering), run the test suite to confirm it passes, and
because this file lives in a protected agentkit directory, include the change in
your PR and request CODEOWNERS approval before merging.
In @.agentkit/spec/commands.yaml:
- Around line 711-738: The --generate-plans flag is currently defaulting to
true, making plan scaffolding run unless explicitly disabled; change the flag
definition for --generate-plans in commands.yaml so its default is false (and
keep required: false) and update the flag description to state it is opt-in
(e.g., "opt-in: scaffold plan files only when provided"); this ensures that
command behavior is non-mutating by default and that scaffolding only occurs
when --generate-plans is explicitly passed.
In @.github/workflows/claude.yml:
- Around line 15-20: The workflow uses a self-hosted runner for
comment-triggered events (see the if condition and the runs-on: self-hosted in
the claude.yml job), which is insecure for public comment triggers; change
runs-on to a GitHub-hosted runner (e.g., ubuntu-latest) or, if you must use
self-hosted, restrict execution by adding strict gatekeeping such as verifying
the actor is trusted (check github.actor in the if expression), limiting to a
dedicated isolated runner group with limited permissions, and disallowing
forks/comments from untrusted actors before allowing the job to run.
In `@scripts/setup-agentkit-branch-governance.sh`:
- Around line 106-114: The for-loop currently iterates "for BRANCH in main main"
which applies branch protection twice; update the loop to list the intended
branches (e.g., "main" only or "main dev"/"main develop") so each branch is
processed once. Locate the loop that sets BRANCH and adjust the second token
(remove the duplicate or replace with the correct branch name), leaving the
existing DRY_RUN check and the gh api --method PUT
"/repos/$REPO/branches/$BRANCH/protection" call intact.
---
Minor comments:
In @.agentkit/templates/scripts/update-changelog.sh:
- Around line 65-67: The PR link in update-changelog.sh builds ENTRY using a
relative path ("../../pull/${PR_NUMBER}") which will break if CHANGELOG.md is at
repo root; update the string construction that uses ENTRY and PR_NUMBER (and
HISTORY_DOC) to use either a root-relative path like "/pull/${PR_NUMBER}" or a
full absolute GitHub URL (e.g.,
"https://github.com/<owner>/<repo>/pull/${PR_NUMBER}") or dynamically derive the
repo URL, so the generated link resolves correctly from the changelog location.
In @.agents/skills/analyze-agents/SKILL.md:
- Around line 37-42: The "Project Context" section has incorrect list nesting
where the "Tech stack" item is indented under "Default branch" (look for the
"Project Context" heading and the list items "Default branch" and "Tech stack");
fix the source spec that produces this generated file so "Tech stack" is a
sibling list item (remove the extra indentation in the template/spec for that
list entry), then re-run the agentkit generation/sync step to regenerate the
markdown output.
In @.github/workflows/issue-label-validation.yml:
- Around line 101-107: The early return guarded by the hasFormFields check
(const hasFormFields = /### (Area|Priority)/i.test(body); if (!hasFormFields) {
core.info(...); return; }) skips label cleanup for edited events; move or invoke
the existing cleanup block that removes area:*, priority:*, severity:*, and
invalid-fields labels to run before this guard (or call it in the no-form
branch) so that stale labels are always cleared when hasFormFields is false;
apply the same change to the duplicate block around lines 152-168 so both
occurrences call the cleanup routine prior to returning.
In @.windsurf/rules/team-strategic-ops.md:
- Line 8: The glob pattern contains a markdown-escaped asterisk (`**/\*`) which
will break Windsurf Cascade matching; update the pattern string to use an
unescaped asterisk (`**/*`) so it matches directories/files correctly (look for
the `**/\*` token in the rule and replace it with `**/*`).
In `@docs/planning/README.md`:
- Around line 331-337: The listed table entries in docs (e.g.,
.agentkit/docs/getting-started/QUICK_START.md, COMMAND_REFERENCE.md,
AGENTS_REFERENCE.md, AGENTS_VS_TEAMS.md, WORKFLOWS.md) are generated and should
not be edited directly; instead update the corresponding source spec/template
under .agentkit/spec/ (modify the spec that generates the affected section or
remove the outdated "In-Flight" entries) and then re-generate the docs with pnpm
-C .agentkit agentkit:sync so the changes persist and CI won't be broken.
In `@scripts/resolve-merge.ps1`:
- Around line 1-7: The generated resolve-merge.ps1 contains non-ASCII em dash
characters causing PowerShell 5.1 encoding issues; fix the AgentKit Forge
upstream template that produces this file by either (a) changing the template to
emit a UTF‑8 BOM on generation, or (b) replacing the em dashes in the template
comment headers with ASCII hyphens or plain text equivalents, then re-run the
AgentKit sync/regeneration step so resolve-merge.ps1 is regenerated without the
non-ASCII characters.
In `@scripts/setup-agentkit-branch-governance.sh`:
- Around line 117-120: The script prints the same verification command twice for
the main branch (the duplicate echo of "gh api
/repos/$REPO/branches/main/protection"); remove the repeated line or replace the
second echo with the intended branch(s) to verify (e.g., change to another
branch name or loop over a BRANCHES variable) so the output only shows the
correct set of verification commands for protected branches.
In `@scripts/sync-split-pr.ps1`:
- Around line 42-48: The generated script sync-split-pr.ps1 emits a local
timestamp with a literal "Z"; fix the source spec files under .agentkit/spec by
replacing the Get-Date format that appends a literal Z with a format that uses
the .NET "K" time zone specifier so timestamps truly represent UTC/offsets,
update all three occurrences in the spec, then regenerate the script using pnpm
-C .agentkit agentkit:sync to produce corrected sync-split-pr.ps1.
In `@scripts/update-changelog.ps1`:
- Around line 1-3: The UTF-8 file header uses a non-ASCII em dash which breaks
PowerShell 5.1; fix by either replacing the em dash with a plain ASCII hyphen in
the generated banner comment lines of scripts/update-changelog.ps1 and
scripts/analyze-agents.ps1, or configure the AgentKit Forge generator to emit a
UTF-8 BOM for generated .ps1 files (update the generator config that produces
the banner in .agentkit/spec or .agentkit/overlays/agentkit-forge) so PowerShell
can read the non-ASCII character.
---
Nitpick comments:
In @.agentkit/engines/node/src/__tests__/discover.test.mjs:
- Line 59: The test description for the spec in the it('returns a discovery
report with expected structure', async () => { ... }) does not follow the repo
convention; rename the string to start with "should" and describe expected
behaviour in plain English (e.g. change to "should return a discovery report
with the expected structure") by updating the it(...) call in the
discover.test.mjs file so the spec reads like "should <verb> <expected outcome>"
while keeping the async test body and assertions unchanged.
In @.agentkit/templates/claude/hooks/protect-templates.ps1:
- Around line 38-40: The empty catch block that swallows parse errors should log
the exception details instead of being silent; inside the catch following the
template parse try (in protect-templates.ps1 / the catch block associated with
the template parsing logic), emit a debug/verbose log including the caught error
($_ or $_.Exception.Message and optionally $_.Exception.StackTrace) via
Write-Verbose or Write-Debug (or Write-Error at Verbose level) so parsing
failures are recorded while still allowing protection to proceed.
In @.agentkit/templates/scripts/sync-split-pr.sh:
- Around line 98-110: The script currently runs git push and gh pr create
without checking failures; add explicit error handling after the git push of
"$NEW_BRANCH" (the git push -u origin "$NEW_BRANCH" command) and after PR
creation (the PR_URL assignment using gh pr create) by testing their exit status
and printing descriptive errors including context (branch name, remote, and hint
about auth/network) and exiting non‑zero on failure; ensure you reference
NEW_BRANCH/COMMIT_MESSAGE in the error output for the commit/push step and
PR_BODY/PR_URL for the PR creation step, e.g., wrap the git push and gh pr
create calls with conditionals or use "|| { echo '...'; exit 1; }" to fail fast
with meaningful messages.
In @.agentkit/templates/scripts/update-changelog.ps1:
- Around line 66-106: Move the embedded JS held in the $NodeScript here-string
into a standalone template module (e.g., templates/scripts/update-changelog.mjs)
and update the .ps1 and .sh callers to run node against that file instead of
embedding the script; specifically, extract the code block that defines
changelogPath/section/entry, file I/O and splice logic into
update-changelog.mjs, then replace the $NodeScript here-string in
update-changelog.ps1 with logic that calls node update-changelog.mjs
<changelogPath> <section> <entry> (and make the same change in the shell
variant), ensuring the new module exports no state and preserves the same CLI
argument behavior and output messages so functions like the existing
file-read/write and console.log/console.error behavior remain identical.
In @.agentkit/templates/scripts/update-changelog.sh:
- Around line 76-77: Sanitize and validate the inputs extracted from
process.argv (const [,, changelogPath, section, entry] = process.argv) before
using them to update the changelog: add checks to trim and reject empty entries,
strip or escape problematic control characters and unescaped markdown sequences
(e.g., backticks, leading hashes, HTML tags) in entry, and validate section
against an allowed set; perform similar minimal validation on changelogPath to
avoid path traversal and ensure it points to the expected file type. Implement
these checks at the top of the Node heredoc (where process.argv is parsed) and
fail early with a clear error when validation fails.
In `@docs/agents/agent-team-matrix.md`:
- Around line 30-48: The docs list many agents with no team mapping (e.g.,
backend, frontend, data, devops, infra, brand-guardian, ui-designer,
content-strategist, growth-analyst, dependency-watcher, environment-manager,
security-auditor, retrospective-analyst, project-shipper, release-manager,
feature-ops); add a short note in the planning docs (or in this
agents/agent-team-matrix.md) stating whether these unassigned agents are
intentionally teamless, planned to be assigned in future work, or will operate
independently, and if applicable add a TODO or link to the roadmap/issue
tracking the future assignment for each category.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 46cea574-c149-44d2-9728-c030c6153b30
⛔ Files ignored due to path filters (1)
pr72-ubuntu-failed.logis excluded by!**/*.log
📒 Files selected for processing (98)
.agentkit/engines/node/src/__tests__/agent-analysis.test.mjs.agentkit/engines/node/src/__tests__/cli.test.mjs.agentkit/engines/node/src/__tests__/discover.test.mjs.agentkit/engines/node/src/agent-analysis.mjs.agentkit/engines/node/src/cli.mjs.agentkit/engines/node/src/synchronize.mjs.agentkit/spec/agents.yaml.agentkit/spec/commands.yaml.agentkit/spec/teams.yaml.agentkit/templates/claude/hooks/protect-templates.ps1.agentkit/templates/claude/hooks/protect-templates.sh.agentkit/templates/scripts/analyze-agents.ps1.agentkit/templates/scripts/analyze-agents.sh.agentkit/templates/scripts/check-documentation-requirement.sh.agentkit/templates/scripts/create-doc.ps1.agentkit/templates/scripts/create-doc.sh.agentkit/templates/scripts/resolve-merge.ps1.agentkit/templates/scripts/resolve-merge.sh.agentkit/templates/scripts/setup-agentkit-branch-governance.ps1.agentkit/templates/scripts/setup-agentkit-branch-governance.sh.agentkit/templates/scripts/sync-issues.sh.agentkit/templates/scripts/sync-split-pr.ps1.agentkit/templates/scripts/sync-split-pr.sh.agentkit/templates/scripts/update-changelog.ps1.agentkit/templates/scripts/update-changelog.sh.agentkit/templates/scripts/validate-documentation.sh.agentkit/templates/scripts/validate-numbering.sh.agents/skills/analyze-agents/SKILL.md.claude/agents/adoption-strategist.md.claude/agents/flow-designer.md.claude/agents/governance-advisor.md.claude/agents/impact-assessor.md.claude/agents/input-clarifier.md.claude/agents/mission-definer.md.claude/agents/portfolio-analyst.md.claude/agents/prompt-engineer.md.claude/agents/release-coordinator.md.claude/agents/role-architect.md.claude/agents/team-validator.md.claude/commands/team-forge.md.claude/commands/team-strategic-ops.md.claude/hooks/protect-templates.ps1.claude/hooks/protect-templates.sh.claude/skills/analyze-agents/SKILL.md.cursor/commands/analyze-agents.md.cursor/rules/team-forge.mdc.cursor/rules/team-strategic-ops.mdc.factory/settings.json.github/agents/adoption-strategist.agent.md.github/agents/flow-designer.agent.md.github/agents/governance-advisor.agent.md.github/agents/impact-assessor.agent.md.github/agents/input-clarifier.agent.md.github/agents/mission-definer.agent.md.github/agents/portfolio-analyst.agent.md.github/agents/prompt-engineer.agent.md.github/agents/release-coordinator.agent.md.github/agents/role-architect.agent.md.github/agents/team-validator.agent.md.github/chatmodes/team-forge.chatmode.md.github/chatmodes/team-strategic-ops.chatmode.md.github/prompts/analyze-agents.prompt.md.github/workflows/claude.yml.github/workflows/issue-label-validation.yml.windsurf/commands/analyze-agents.md.windsurf/commands/sync-backlog.md.windsurf/rules/team-forge.md.windsurf/rules/team-strategic-ops.mdAGENT_TEAMS.mdCHANGELOG.md__tests__/task-types.test.mjsdocs/agents/agent-team-matrix.mddocs/planning/BRANCH_PLAN.mddocs/planning/README.mddocs/planning/agents-teams/forge-team.mddocs/planning/agents-teams/strategic-ops-team.mddocs/planning/brand/brand-enhancements.mddocs/planning/bugs/dx-improvements.mddocs/planning/bugs/prd-detector-path.mddocs/planning/cli-packaging/cli-team-proposal.mddocs/planning/intake/intake-agent-proposal.mdscripts/analyze-agents.ps1scripts/analyze-agents.shscripts/check-documentation-requirement.shscripts/create-doc.ps1scripts/create-doc.shscripts/create-quality-issues.shscripts/resolve-merge.ps1scripts/resolve-merge.shscripts/setup-agentkit-branch-governance.ps1scripts/setup-agentkit-branch-governance.shscripts/sync-issues.shscripts/sync-split-pr.ps1scripts/sync-split-pr.shscripts/update-changelog.ps1scripts/update-changelog.shscripts/validate-documentation.shscripts/validate-numbering.sh
💤 Files with no reviewable changes (1)
- tests/task-types.test.mjs
…cket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Contributor
There was a problem hiding this comment.
Actionable comments posted: 12
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
.agentkit/templates/claude/hooks/protect-templates.sh (1)
36-50:⚠️ Potential issue | 🟠 MajorDon't recommend an action this hook immediately forbids.
.agentkit/spec/is still inPROTECTED_PATTERNS, so the denial text sends the agent into a dead end: edit.agentkit/spec/→ denied again. Reword this to tell a human contributor to update the spec and rerun sync, or stop protecting.agentkit/spec/here.Suggested wording
-# repo, or modify .agentkit/spec/ and run sync to regenerate. +# repo. For downstream project configuration changes, ask a human contributor +# to update .agentkit/spec/ and run sync to regenerate. ... - --arg reason "Blocked: '${FILE_PATH}' is an AgentKit Forge source file. These files are the upstream source-of-truth and must not be modified directly by AI agents. To propose changes, create a PR to the agentkit-forge repository targeting the relevant spec or template. If you need to change project configuration, edit the YAML specs in .agentkit/spec/ and run 'pnpm -C .agentkit agentkit:sync'." \ + --arg reason "Blocked: '${FILE_PATH}' is an AgentKit Forge source file. These files are the upstream source-of-truth and must not be modified directly by AI agents. To propose template or engine changes, create a PR to the agentkit-forge repository. For downstream project configuration changes, ask a human contributor to update the YAML specs in .agentkit/spec/ and run 'pnpm -C .agentkit agentkit:sync'." \Based on learnings: Users (not AI agents) may modify
.agentkit/spec/YAML files and runagentkit sync.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/claude/hooks/protect-templates.sh around lines 36 - 50, The denial message currently blocks edits to '.agentkit/spec/' while that path remains in PROTECTED_PATTERNS, creating a dead end; update the PROTECTED_PATTERNS array or the jq --arg reason text so it no longer tells users to perform the exact forbidden action. Specifically, either remove the '\.agentkit/spec/' entry from PROTECTED_PATTERNS or change the jq --arg reason string built in protect-templates.sh to explicitly tell human contributors they may edit .agentkit/spec/ YAML and then run 'pnpm -C .agentkit agentkit:sync' (while still noting AI agents must not modify forge source files), so the message refers to a viable human workflow instead of implying an action that will be blocked.
♻️ Duplicate comments (1)
.agentkit/templates/scripts/update-changelog.ps1 (1)
108-113:⚠️ Potential issue | 🔴 CriticalFile is protected infrastructure code and cannot be directly modified per coding guidelines.
This file is in
.agentkit/templates/which is part of the upstream sync engine — it must not be modified directly by AI agents. A PreToolUse hook enforces this at runtime.The technical concerns about temp file orphaning and missing error handling for
nodeexit codes are valid:
[System.IO.Path]::GetTempFileName()creates a 0-byte.tmpfile on disk; appending.cjsleaves that original file orphaned.- The script never checks
$LASTEXITCODE, and PowerShell only raises errors on non-zero native-command exits when$PSNativeCommandUseErrorActionPreferenceis enabled.However, fixes to this template must be coordinated through the AgentKit Forge upstream repository, not applied directly to
.agentkit/templates/in this repository.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/scripts/update-changelog.ps1 around lines 108 - 113, This protected template must not be edited here; instead open a patch in the AgentKit Forge upstream repo to: stop using [System.IO.Path]::GetTempFileName() + '.cjs' (which leaves the zero-byte tmp orphan) and create the temp script atomically (e.g., New-Item with a GUID-based filename in [System.IO.Path]::GetTempPath()), ensure both the original zero-byte name (if any) and the created .cjs are removed in the finally block, and add explicit native-exit handling after invoking node (check $LASTEXITCODE or capture Start-Process -Wait -PassThru and throw/exit on non-zero) so failures are propagated; refer to the variables/methods $TmpScript, $NodeScript, $Changelog, $Section, $Entry, GetTempFileName, New-Item, Remove-Item, and $LASTEXITCODE when preparing the upstream change.
🧹 Nitpick comments (3)
.agentkit/templates/claude/hooks/protect-templates.ps1 (1)
38-40: Empty catch block flagged by PSScriptAnalyzer.The intent to silently proceed with protection on parse errors is reasonable, but the empty catch block triggers a linter warning. Consider adding a minimal acknowledgment.
♻️ Suggested improvement
} catch { - # Ignore parse errors, proceed with protection + # Ignore parse errors, proceed with protection (safe fallback) + $null = $_.Exception }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/claude/hooks/protect-templates.ps1 around lines 38 - 40, The empty catch block in protect-templates.ps1 (the catch { ... } after the parse attempt) triggers PSScriptAnalyzer; update that catch to include a minimal acknowledgment such as logging or discarding the exception so the linter is satisfied—e.g., call Write-Verbose or Write-Verbose "Parse error ignored: $_" or assign $null = $_—while keeping the original behavior of proceeding with protection..agentkit/spec/commands.yaml (1)
735-738: Consider defaulting--generate-planstofalse.The
--generate-plansflag defaults totrue, which means running/project-reviewwill automatically scaffold plan files for all critical/high findings intodocs/planning/review-findings/. This could generate many files unexpectedly on first run.Consider defaulting to
falseto make plan generation opt-in, allowing users to review findings first before committing to plan file creation.💡 Proposed change
- name: --generate-plans description: 'After Phase 2, scaffold plan files from critical/high findings into docs/planning/review-findings/' type: boolean - default: true + default: false🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/spec/commands.yaml around lines 735 - 738, Change the boolean default for the CLI flag --generate-plans from true to false so plan scaffolding is opt-in; update the flag definition where "name: --generate-plans" and "default: true" are declared to use "default: false" and, if helpful, adjust the description to note that plan generation is optional and must be explicitly enabled by the user..claude/commands/team-cost-ops.md (1)
96-103: Lock acquisition and renewal strategy is well-designed.The locking mechanism with atomic directory creation (
mkdir), TTL-based lease expiration, stale-lock takeover via atomic claim, and bounded retry with jittered backoff provides robust concurrency control. The explicit handling of lock renewal failures with rollback strategy (lines 104-121) is thorough.One consideration: the
lock_ttl_minutesdefault of 60 minutes (line 97) may be long for typical agent tasks. If tasks typically complete in under 10 minutes, a shorter default (e.g., 15-30 minutes) would reduce stale-lock wait times.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/team-cost-ops.md around lines 96 - 103, The default lock_ttl_minutes is too long (60) for typical short-running agent tasks; change the default to a shorter value (e.g., 15–30 minutes) by updating the lock_ttl_minutes default, recomputing ttlMs = lock_ttl_minutes * 60_000 accordingly, and adjust any documentation/comments referencing the 60-minute default; ensure the change is configurable so operators can tune lock_ttl_minutes for longer tasks and update any tests or examples that assume a 60-minute TTL.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In @.agentkit/spec/agents.yaml:
- Around line 1640-1717: The two agents vendor-arbitrage-analyst and
grant-hunter are declared owners of docs under docs/cost-ops/** but their
preferred-tools lists only include read/search/web tools, so they cannot update
the artifacts; update both agent blocks (the preferred-tools arrays for id:
vendor-arbitrage-analyst and id: grant-hunter) to include a write-capable tool
(for example add "Write" or your project's editor tool) so they can perform
document tasks like updating reports, calendars and pipelines.
- Around line 725-758: The spec-compliance-auditor cannot write reports because
its focus globs and tools are read-only; update the agent spec by adding the
output directory to its focus (e.g., add 'docs/agents/**' to the focus list) and
grant write capability by adding 'Write' to preferred-tools (or replace 'Read'
with 'Read' and 'Write'); keep the existing domain-rule referencing docs/agents/
so the agent can legally place generated scorecards there.
In @.agentkit/spec/rules.yaml:
- Around line 1383-1389: The applies-to glob "docs/planning/cost/**" in the
applies-to list does not match the new planning docs under
"docs/planning/cost-governance/**"; update the pattern in the applies-to block
(the list containing 'docs/cost-ops/**', 'docs/planning/cost/**', etc.) to
include the correct path(s) — for example replace or add a pattern such as
'docs/planning/**' or specifically 'docs/planning/cost-governance/**' so the
ai-cost-ops rules will apply to the new documents.
- Around line 1377-1391: The added "phase" key in the domain block is invalid
under ruleDomainSchema; remove the phase: [planning, implementation] entry from
the domain block in rules.yaml (or relocate it to a permitted place such as
inside a convention or rule-level entry if that was intended) so the domain
object only contains domain, description, applies-to, and conventions, and
re-run spec validation against ruleDomainSchema.
In @.agentkit/templates/claude/hooks/protect-templates.ps1:
- Around line 29-41: The relative path in the $agentkitPkg assignment is wrong
and only climbs two levels; update the Join-Path call that sets $agentkitPkg so
it climbs three parent directories from $scriptDir to reach the repository root
.agentkit/package.json (i.e., use a relative path with three ".." segments
instead of two), so the check that reads and parses $agentkitPkg (the
Get-Content | ConvertFrom-Json block and subsequent $pkg.name ===
'agentkit-forge-runtime' check) correctly detects the source repo and exits.
In @.agentkit/templates/claude/hooks/protect-templates.sh:
- Around line 20-31: The protection script detects the source repo via
AGENTKIT_PKG / PKG_NAME (checks for "agentkit-forge-runtime" and exits early)
but the protection error message shown when PROTECTED_PATTERNS (which includes
".agentkit/spec/") is violated still tells users to "edit the YAML specs in
.agentkit/spec/ and run 'pnpm -C .agentkit agentkit:sync'", which is misleading
for downstream repos; update the error/echo text in protect-templates.sh (the
block that enforces PROTECTED_PATTERNS and prints the rejection message) to be
context-aware: if PKG_NAME == "agentkit-forge-runtime" allow the sync
instruction, otherwise clearly state that edits to .agentkit/spec/ must be made
via a PR to the agentkit-forge repo (or contact maintainers) and do not instruct
running the local pnpm sync command for downstream repos.
In @.agentkit/templates/scripts/update-changelog.ps1:
- Around line 92-99: The loop that advances insertAt (using variables insertAt,
sectionIdx, blockEnd and the lines[] array) doesn't stop at the '---' divider so
a new bullet can be placed after the divider; update the while condition to also
break when the current line is the divider (e.g., when lines[insertAt].trim()
=== '---' or lines[insertAt].startsWith('---')) so the loop stops before the
divider and the new item is appended inside the existing [Unreleased] section.
In @.github/instructions/languages/ai-cost-ops.md:
- Around line 7-10: Add a missing "Applies To" section (with the same file
pattern list used in the Claude variant) immediately above the existing "##
Advisory Rules" heading; include these patterns as separate lines:
docs/cost-ops/**, docs/planning/cost/**, config/models/**, config/pricing/**,
**/*.prompt, **/*.prompt.md; also add a short note advising contributors to
submit a PR to the agentkit-forge repo when changes to templates, specs, or
engines are required.
In `@AGENT_BACKLOG.md`:
- Around line 155-157: The AGENT_BACKLOG entry shows T13-Cost-Ops →
T12-Strategic ("Provider cost visibility") but the T13-Cost-Ops handoff-chain in
teams.yaml only lists [infra, product]; either remove the T13-Cost-Ops →
T12-Strategic dependency from the backlog or add "strategic-ops" to the
T13-Cost-Ops handoff-chain in teams.yaml so the backlog and handoff-chain are
consistent (refer to the T13-Cost-Ops, T12-Strategic, and the "Provider cost
visibility" dependency).
In `@AGENT_TEAMS.md`:
- Line 38: The COST OPS row in the table only references `docs/planning/cost/**`
but your new artifacts live under `docs/planning/cost-governance/**`, so update
the source spec in `.agentkit/spec/` (not the generated `AGENT_TEAMS.md`) to
include `docs/planning/cost-governance/**` for the COST OPS team and then
regenerate docs by running `pnpm -C .agentkit agentkit:sync`; specifically edit
the COST OPS spec entry in `.agentkit/spec/*` to widen the planning scope, save,
and run the sync command so the generated `AGENT_TEAMS.md` reflects the change.
In `@docs/planning/cost-governance/multi-provider-infra-cost.md`:
- Around line 69-70: Replace the proposed "attribute AI spend via provider tags"
approach for the /cost-centres feature with a versioned reference-table
ownership model: add a maintained mapping table keyed by provider+model+workload
that resolves to a cost centre (and version metadata), change the /cost-centres
design to consult that mapping when attributing AI spend (falling back to
"<unassigned>" for unmapped resources), and remove/tag-as-deprecated any logic
or documentation that relies on provider tags for ownership so all attribution
uses the reference table instead.
In `@docs/planning/cost-governance/web-intake-expansion.md`:
- Around line 1-115: This planning doc titled "Cost Ops Web Intake & Crawler
Expansion" is outside the repository's canonical docs taxonomy; move the
markdown into the appropriate docs category (i.e., under the repo's
domain-driven docs tree) or update the project's canonical taxonomy/index to
include this new subtree, and ensure the doc's frontmatter/metadata and the
documentation sidebar/TOC are updated so it appears in navigation; verify
internal links and any references to the "Cost Ops Web Intake & Crawler
Expansion" doc are updated accordingly.
---
Outside diff comments:
In @.agentkit/templates/claude/hooks/protect-templates.sh:
- Around line 36-50: The denial message currently blocks edits to
'.agentkit/spec/' while that path remains in PROTECTED_PATTERNS, creating a dead
end; update the PROTECTED_PATTERNS array or the jq --arg reason text so it no
longer tells users to perform the exact forbidden action. Specifically, either
remove the '\.agentkit/spec/' entry from PROTECTED_PATTERNS or change the jq
--arg reason string built in protect-templates.sh to explicitly tell human
contributors they may edit .agentkit/spec/ YAML and then run 'pnpm -C .agentkit
agentkit:sync' (while still noting AI agents must not modify forge source
files), so the message refers to a viable human workflow instead of implying an
action that will be blocked.
---
Duplicate comments:
In @.agentkit/templates/scripts/update-changelog.ps1:
- Around line 108-113: This protected template must not be edited here; instead
open a patch in the AgentKit Forge upstream repo to: stop using
[System.IO.Path]::GetTempFileName() + '.cjs' (which leaves the zero-byte tmp
orphan) and create the temp script atomically (e.g., New-Item with a GUID-based
filename in [System.IO.Path]::GetTempPath()), ensure both the original zero-byte
name (if any) and the created .cjs are removed in the finally block, and add
explicit native-exit handling after invoking node (check $LASTEXITCODE or
capture Start-Process -Wait -PassThru and throw/exit on non-zero) so failures
are propagated; refer to the variables/methods $TmpScript, $NodeScript,
$Changelog, $Section, $Entry, GetTempFileName, New-Item, Remove-Item, and
$LASTEXITCODE when preparing the upstream change.
---
Nitpick comments:
In @.agentkit/spec/commands.yaml:
- Around line 735-738: Change the boolean default for the CLI flag
--generate-plans from true to false so plan scaffolding is opt-in; update the
flag definition where "name: --generate-plans" and "default: true" are declared
to use "default: false" and, if helpful, adjust the description to note that
plan generation is optional and must be explicitly enabled by the user.
In @.agentkit/templates/claude/hooks/protect-templates.ps1:
- Around line 38-40: The empty catch block in protect-templates.ps1 (the catch {
... } after the parse attempt) triggers PSScriptAnalyzer; update that catch to
include a minimal acknowledgment such as logging or discarding the exception so
the linter is satisfied—e.g., call Write-Verbose or Write-Verbose "Parse error
ignored: $_" or assign $null = $_—while keeping the original behavior of
proceeding with protection.
In @.claude/commands/team-cost-ops.md:
- Around line 96-103: The default lock_ttl_minutes is too long (60) for typical
short-running agent tasks; change the default to a shorter value (e.g., 15–30
minutes) by updating the lock_ttl_minutes default, recomputing ttlMs =
lock_ttl_minutes * 60_000 accordingly, and adjust any documentation/comments
referencing the 60-minute default; ensure the change is configurable so
operators can tune lock_ttl_minutes for longer tasks and update any tests or
examples that assume a 60-minute TTL.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 9181b37c-7547-41be-88d9-4cbed5f75b4f
📒 Files selected for processing (39)
.agentkit/spec/agents.yaml.agentkit/spec/commands.yaml.agentkit/spec/rules.yaml.agentkit/spec/teams.yaml.agentkit/templates/claude/hooks/protect-templates.ps1.agentkit/templates/claude/hooks/protect-templates.sh.agentkit/templates/scripts/resolve-merge.sh.agentkit/templates/scripts/update-changelog.ps1.claude/agents/cost-ops-monitor.md.claude/agents/grant-hunter.md.claude/agents/model-economist.md.claude/agents/spec-compliance-auditor.md.claude/agents/token-efficiency-engineer.md.claude/agents/vendor-arbitrage-analyst.md.claude/commands/team-cost-ops.md.claude/hooks/protect-templates.ps1.claude/hooks/protect-templates.sh.claude/rules/languages/ai-cost-ops.md.clinerules/ai-cost-ops.md.clinerules/languages/ai-cost-ops.md.cursor/rules/languages/ai-cost-ops.md.cursor/rules/team-cost-ops.mdc.github/agents/cost-ops-monitor.agent.md.github/agents/grant-hunter.agent.md.github/agents/model-economist.agent.md.github/agents/spec-compliance-auditor.agent.md.github/agents/token-efficiency-engineer.agent.md.github/agents/vendor-arbitrage-analyst.agent.md.github/chatmodes/team-cost-ops.chatmode.md.github/instructions/languages/ai-cost-ops.md.roo/rules/ai-cost-ops.md.roo/rules/languages/ai-cost-ops.md.windsurf/commands/sync-backlog.md.windsurf/rules/languages/ai-cost-ops.md.windsurf/rules/team-cost-ops.mdAGENT_BACKLOG.mdAGENT_TEAMS.mddocs/planning/cost-governance/multi-provider-infra-cost.mddocs/planning/cost-governance/web-intake-expansion.md
✅ Files skipped from review due to trivial changes (6)
- .claude/agents/spec-compliance-auditor.md
- .clinerules/languages/ai-cost-ops.md
- .github/agents/cost-ops-monitor.agent.md
- .roo/rules/ai-cost-ops.md
- .windsurf/rules/languages/ai-cost-ops.md
- .claude/agents/vendor-arbitrage-analyst.md
🚧 Files skipped from review as they are similar to previous changes (1)
- .agentkit/templates/scripts/resolve-merge.sh
JustAGhosT
changed the title
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Collaborator
Author
All CodeRabbit Review Comments AddressedCommit \ addresses all 29 review comments (6 critical, 17 major, 4 minor, 2 advisory): Critical fixes:
Major fixes:
Minor/Advisory:
|
Contributor
There was a problem hiding this comment.
Actionable comments posted: 11
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (3)
.agentkit/templates/claude/hooks/protect-templates.sh (1)
39-48:⚠️ Potential issue | 🟠 MajorNormalize
file_pathbefore applying protected-path regexes.The current patterns only match forward slashes, so Windows-style paths can bypass the protection and allow writes into
.agentkit/templates/,.agentkit/engines/,.agentkit/overlays/, and.agentkit/bin/. NormalizeFILE_PATHfirst, or match both separators.Suggested fix
+NORMALIZED_FILE_PATH="${FILE_PATH//\\//}" + for pattern in "${PROTECTED_PATTERNS[@]}"; do - if echo "$FILE_PATH" | grep -qE "$pattern"; then + if echo "$NORMALIZED_FILE_PATH" | grep -qE "$pattern"; then jq -n \As per coding guidelines, "AI agents must never directly modify files in .agentkit/templates/, .agentkit/engines/, .agentkit/overlays/, or .agentkit/bin/. These directories contain the upstream sync engine, templates, and CLI scripts. A PreToolUse hook enforces this at runtime."
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/claude/hooks/protect-templates.sh around lines 39 - 48, Normalize FILE_PATH before testing against PROTECTED_PATTERNS so Windows backslashes can't bypass the checks: in protect-templates.sh, before the for loop that iterates over PROTECTED_PATTERNS, transform FILE_PATH to a normalized variable (e.g., replace backslashes with forward slashes) and use that normalized path in the grep -qE check; alternatively, update PROTECTED_PATTERNS to accept both '/' and '\' separators, but ensure the script uses the normalized variable (refer to FILE_PATH, PROTECTED_PATTERNS and the for pattern loop) when matching.docs/planning/intake/intake-agent-proposal.md (1)
86-90:⚠️ Potential issue | 🟡 MinorAgent Definition focus paths still use non-canonical locations.
The table in "Supported Document Types" was updated to use canonical paths (
docs/product/,docs/orchestration/), but the Agent Definition'sfocusblock still references the old non-standard paths:
docs/design/**→ should bedocs/product/**ordocs/architecture/**docs/processes/**→ should bedocs/orchestration/**docs/prd/**→ should bedocs/product/**(PRDs live under product)These paths should be aligned with the canonical taxonomy to ensure the intake agent watches the correct directories.
As per coding guidelines, "All project documentation must follow the domain-driven structure under docs/. Categories: product (vision, strategy, personas), architecture/ (specs, decisions, diagrams), orchestration (guides, protocols)..."
🔧 Suggested fix
focus: - 'docs/product/**' - 'docs/architecture/specs/**' - - 'docs/design/**' - - 'docs/processes/**' - - 'docs/prd/**' + - 'docs/orchestration/**' - 'AGENT_BACKLOG.md'🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@docs/planning/intake/intake-agent-proposal.md` around lines 86 - 90, Update the Agent Definition's focus block so it uses the canonical docs taxonomy instead of legacy paths: replace any occurrences of "docs/design/**" with "docs/product/**" or "docs/architecture/**" as appropriate, replace "docs/processes/**" with "docs/orchestration/**", and replace "docs/prd/**" with "docs/product/**"; ensure these changes are made within the Agent Definition's focus block so the intake agent will watch the correct directories..claude/hooks/protect-templates.sh (1)
42-53:⚠️ Potential issue | 🟠 MajorDon't send users to a path this hook also blocks.
\.agentkit/spec/is still listed inPROTECTED_PATTERNS, so the denial text points downstream users to a location the nextWrite/Editwill reject again. Please either exempt local specs or change the guidance to the actual writable configuration path.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/hooks/protect-templates.sh around lines 42 - 53, The deny message references .agentkit/spec/ while that path is still blocked in PROTECTED_PATTERNS, causing a loop; either remove the '\.agentkit/spec/' entry from PROTECTED_PATTERNS to allow local spec edits, or keep it blocked but update the jq --arg reason string used where FILE_PATH matches to point to a writable configuration location (for example an overlays or project config path) and include correct instructions (e.g., run 'pnpm -C .agentkit agentkit:sync' only if that location is writable); modify the PROTECTED_PATTERNS array or the --arg reason text accordingly so the guidance does not point to a blocked path and reference PROTECTED_PATTERNS, FILE_PATH and the jq --arg reason usage to locate the exact places to change.
♻️ Duplicate comments (5)
.agentkit/templates/claude/hooks/protect-templates.sh (1)
49-50:⚠️ Potential issue | 🟠 MajorMake the downstream denial message consistent with the actual block.
Line 50 still tells downstream users to edit
.agentkit/spec/locally and runpnpm -C .agentkit agentkit:sync, but this hook blocks that path in the same repo. Please split the message by repo type so downstream repos are only told to open a PR againstagentkit-forge.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/claude/hooks/protect-templates.sh around lines 49 - 50, Update the denial message in protect-templates.sh so it varies by repo type: in the jq --arg reason that references '${FILE_PATH}', detect whether the current repo is the upstream agentkit-forge (e.g., check the git remote URL or an env flag used in this script) and, if it is, keep the existing guidance about opening a PR to agentkit-forge; otherwise (downstream repos) replace the local-edit/sync instructions with a single instruction telling users to propose changes by opening a PR against the agentkit-forge repository. Locate the message construction around the jq --arg reason call in protect-templates.sh and implement a simple if/else that sets the reason string accordingly..agentkit/templates/scripts/update-changelog.ps1 (1)
88-99:⚠️ Potential issue | 🟡 MinorStill missing
---divider check in the while loop.The loop that finds the insertion point (lines 90-97) doesn't stop at a
---divider. If the last populated section is followed directly by the divider, the new bullet will be inserted after the divider instead of inside the section.🐛 Proposed fix
let insertAt = sectionIdx + 1; while ( insertAt < blockEnd && lines[insertAt].trim() !== '' && + lines[insertAt].trim() !== '---' && !lines[insertAt].startsWith('###') && !lines[insertAt].startsWith('##') ) { insertAt++; }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/scripts/update-changelog.ps1 around lines 88 - 99, The insertion loop in update-changelog.ps1 (variables: insertAt, sectionIdx, blockEnd, lines, entry) doesn't stop at a horizontal divider, so add a check to break when the current line is the divider (e.g., lines[insertAt].trim() === '---' or lines[insertAt].startsWith('---')) alongside the existing blank/heading checks; update the while condition to include "!lines[insertAt].trim().startsWith('---')" (or equivalent) so the new entry is inserted before the divider and inside the section..agentkit/templates/scripts/resolve-merge.sh (1)
79-89:⚠️ Potential issue | 🟠 MajorPattern matching is broken for glob-style patterns.
The patterns in
GENERATED_PATTERNSuse glob syntax (e.g.,.agents/skills/*/SKILL.md), but line 88 usesgrep -Ewhich expects regex. In regex:
.matches any character (should be\.for literal dot)*means "zero or more of preceding character" (not a wildcard)For example,
.agents/skills/*/SKILL.mdwon't matchagents/skills/foo/SKILL.mdcorrectly.The previous review suggested using shell glob matching instead.
🔧 Suggested fix using shell glob matching
for pattern in "${GENERATED_PATTERNS[@]}"; do - # Find conflicted files matching this pattern while IFS= read -r file; do [ -z "$file" ] && continue + # Use shell glob matching instead of grep + # shellcheck disable=SC2053 + [[ "$file" == $pattern ]] || continue if git checkout --theirs -- "$file" 2>/dev/null; then git add "$file" 2>/dev/null ok "Auto-resolved (accept upstream): $file" auto_resolved=$((auto_resolved + 1)) fi - done < <(git diff --name-only --diff-filter=U | grep -E "$pattern" 2>/dev/null || true) + done < <(git diff --name-only --diff-filter=U 2>/dev/null || true) done🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/scripts/resolve-merge.sh around lines 79 - 89, The current loop uses grep -E on git diff output which treats GENERATED_PATTERNS (e.g., `.agents/skills/*/SKILL.md`) as regexes and breaks glob matching; update the loop that iterates over GENERATED_PATTERNS so it filters conflicted files using shell glob matching instead of grep -E: read filenames from git diff --name-only --diff-filter=U and for each file test the glob against the pattern using a shell glob comparison (e.g., [[ "$file" == $pattern ]] or case) before running git checkout --theirs / git add; keep the rest of the logic (ok message, auto_resolved increment) intact and remove the grep -E usage..claude/hooks/protect-templates.ps1 (1)
29-39:⚠️ Potential issue | 🟠 MajorThe PowerShell bypass path is still one directory too high.
Line 33 joins
..\..\..\package.jsonfrom.claude/hooks/, which points outside the repo root. The source-repo bypass therefore never matches and the hook keeps blocking writes in the forge repo itself.Suggested fix
$scriptDir = Split-Path -Parent $MyInvocation.MyCommand.Definition -$agentkitPkg = Join-Path $scriptDir '..\..\..\package.json' +$agentkitPkg = Join-Path $scriptDir '..\..\package.json'🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/hooks/protect-templates.ps1 around lines 29 - 39, The relative path used to detect the agentkit-forge source repo is one level too high: update the $agentkitPkg assignment (which currently joins $scriptDir with '..\..\..\package.json') to use one fewer parent segment so it points at the repo root (e.g. '..\..\package.json' relative to $scriptDir), then keep the existing Get-Content/ConvertFrom-Json check against $pkg.name to allow the hook to exit when in the source repo..agentkit/spec/agents.yaml (1)
725-731:⚠️ Potential issue | 🟡 Minor
spec-compliance-auditorfocus is missing output directory.The domain-rules state "place reports in docs/agents/" (line 742), but
docs/agents/**is not included in the focus globs. The agent won't be able to write scorecards and recommendations to its declared output location.Suggested fix
focus: - '.agentkit/spec/agents.yaml' - '.agentkit/spec/teams.yaml' - '.claude/state/tasks/**' - '.claude/state/events.log' - 'docs/history/**' + - 'docs/agents/**'🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/spec/agents.yaml around lines 725 - 731, The spec-compliance-auditor agent's focus list (the YAML key "focus" under the spec for spec-compliance-auditor) is missing the output pattern for its report directory; add "docs/agents/**" to the existing focus array (alongside '.agentkit/spec/agents.yaml', '.agentkit/spec/teams.yaml', '.claude/state/tasks/**', '.claude/state/events.log', 'docs/history/**') so the agent can write its reports into the declared output location referenced by the domain-rules.
🧹 Nitpick comments (8)
.claude/commands/team-cost-ops.md (1)
282-282: Consider standardizing hyphenation of "follow-up".The document uses both "followup" (lines 282, 290) and "follow-up" (line 290 mentions "follow-ups"). For consistency and readability, standardize on either "follow-up" or "followup" throughout the document.
Industry convention typically favors the hyphenated form "follow-up" as a noun/adjective.
Also applies to: 290-290
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.claude/commands/team-cost-ops.md at line 282, Standardize the hyphenation of "follow-up" across the document by replacing all instances of "followup" and "follow-ups" with the hyphenated form "follow-up" (including in strings like the status name used in the orchestrator such as "ready-for-followup" — update that to "ready-for-follow-up" if you decide to hyphenate status names consistently) and any references in handoffContext text; ensure consistency for both noun/adjective uses and for status identifiers like "needs-review" vs "ready-for-follow-up" so the term is uniform throughout.scripts/analyze-agents.ps1 (1)
26-26: Consider usingJoin-Pathfor cross-platform consistency.While forward slashes work on PowerShell, using
Join-Pathwould be consistent with the style elsewhere in the repo's PowerShell scripts.♻️ Optional refactor
-node "$ProjectRoot/.agentkit/engines/node/src/cli.mjs" `@cliArgs` +$cliPath = Join-Path $ProjectRoot '.agentkit' 'engines' 'node' 'src' 'cli.mjs' +node $cliPath `@cliArgs`🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@scripts/analyze-agents.ps1` at line 26, Replace the inline string path to the Node CLI with a Join-Path-built path and invoke it via the call operator so the script is platform-consistent: build the CLI path from $ProjectRoot using Join-Path (targeting the .agentkit/engines/node/src/cli.mjs segments) and then call node with that joined path and the existing `@cliArgs`; reference $ProjectRoot, cli.mjs and `@cliArgs` to locate where to change the invocation..agentkit/templates/scripts/resolve-merge.ps1 (1)
70-89: Consider caching the conflicted file list to avoid repeated git calls.The script calls
git diff --name-only --diff-filter=Umultiple times (lines 71, 101, 118, 133). For efficiency, consider caching the initial list and filtering it in-memory.♻️ Optional optimization
+$allConflicted = git diff --name-only --diff-filter=U 2>$null +if (-not $allConflicted) { $allConflicted = @() } + $autoResolved = 0 -$conflicted = git diff --name-only --diff-filter=U 2>$null -if (-not $conflicted) { $conflicted = @() } +$conflicted = $allConflicted.Clone() foreach ($file in $conflicted) { # ... auto-resolve logic ... + # After resolving, filter out from remaining list }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/scripts/resolve-merge.ps1 around lines 70 - 89, Cache the result of the git conflicted-file query into a single variable (currently $conflicted) once up front and reuse that array for all subsequent operations instead of re-running `git diff --name-only --diff-filter=U` multiple times; modify the code paths that currently re-run the command to filter and operate on the in-memory $conflicted list (e.g., the loops that use foreach ($file in $conflicted) and the checks against $generatedPatterns) so all filtering (generated file pattern matching) and git actions (git checkout --theirs, git add) operate on the cached list..agentkit/templates/scripts/setup-agentkit-branch-governance.ps1 (1)
77-81: Remove unused$branchExistsvariable.The variable is assigned but never used; only
$LASTEXITCODEis checked. Either remove the assignment or use the response content.♻️ Proposed fix
- $branchExists = gh api "/repos/$Repo/branches/$branch" 2>$null + gh api "/repos/$Repo/branches/$branch" 2>$null | Out-Null if ($LASTEXITCODE -ne 0) { Write-Host "[skip] Branch '$branch' does not exist on $Repo — skipping protection." continue }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/scripts/setup-agentkit-branch-governance.ps1 around lines 77 - 81, The assignment to $branchExists is unused; update the gh api call around the branch existence check so you either (A) remove the $branchExists assignment and invoke gh api without capturing output (e.g. discard stdout/stderr) and continue to check $LASTEXITCODE, or (B) capture the API response into a meaningful variable (e.g. $response) and use that response instead of relying solely on $LASTEXITCODE; modify the block referencing $branchExists, $LASTEXITCODE, and the gh api invocation accordingly so no unused variable remains..agentkit/templates/scripts/analyze-agents.ps1 (1)
14-23: Consider adding a preflight check fornode.The script assumes
nodeis available but doesn't validate. Adding a check similar to other scripts in this PR would improve error messages.♻️ Optional preflight check
$ErrorActionPreference = 'Stop' + +if (-not (Get-Command node -ErrorAction SilentlyContinue)) { + Write-Error "Node.js is not installed or not in PATH." + exit 1 +} + $ScriptDir = Split-Path -Parent $MyInvocation.MyCommand.Path🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/scripts/analyze-agents.ps1 around lines 14 - 23, Add a preflight check that verifies the `node` executable is available before invoking it: use a command check (e.g., Get-Command node or equivalent) early in the script (before the node "$ProjectRoot/.agentkit/engines/node/src/cli.mjs" call) and if not found write a clear error message and exit with non-zero status; modify the script surrounding the $ProjectRoot, $cliArgs and node invocation to bail out with a descriptive message when `node` is missing so the subsequent node call is never attempted..agentkit/templates/scripts/create-doc.ps1 (1)
35-36: Consider using named parameters forJoin-Pathwith multiple path segments.PSScriptAnalyzer flags positional parameters. While functional, named parameters improve readability for multi-segment joins.
♻️ Optional refactor
-$HistoryDir = Join-Path $RepoRoot 'docs' 'history' +$HistoryDir = Join-Path -Path $RepoRoot -ChildPath 'docs' | Join-Path -ChildPath 'history'Or keep as-is since multi-argument
Join-Pathis valid in PowerShell 6+.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.agentkit/templates/scripts/create-doc.ps1 around lines 35 - 36, Replace positional arguments to Join-Path with named parameters to improve readability: update the $HistoryDir assignment to call Join-Path -Path $RepoRoot -ChildPath 'docs' and then Join-Path -Path (that result) -ChildPath 'history' (or use nested Join-Path calls), and update $IndexFile to call Join-Path -Path $HistoryDir -ChildPath '.index.json'; reference $HistoryDir, $IndexFile and Join-Path when making these changes..agentkit/engines/node/src/cli.mjs (1)
161-216: Self-validation warning inloadCommandFlagsis helpful.The loop at lines 204-213 that warns about flags missing type definitions is a good defensive measure. However, consider whether this should be a hard error in CI/production rather than just a warning.
.agentkit/engines/node/src/agent-analysis.mjs (1)
541-569: Potential memoization issue incomputeAgentReachabilitywith cycles.The
visitedset is passed by reference to detect cycles, but results are memoized after recursive calls complete. For graphs with cycles, an agent visited mid-recursion returns an empty set (line 546), which gets added to the parent's result. The parent then caches this potentially incomplete result.For the current use case (depends-on graphs which should be DAGs), this is likely fine. If cycles exist,
detectCircularDepsshould catch them separately.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In @.agentkit/templates/scripts/sync-split-pr.ps1:
- Around line 13-17: Add an authentication check using "gh auth status"
immediately after the existing Get-Command gh check: run "gh auth status" and if
it reports unauthenticated, emit an error and exit to prevent later side
effects; also ensure the same verification is present before the
branch/commit/push/PR sequence referenced by the branch creation/commit/push
logic and the gh pr create invocation so those operations only run when gh is
authenticated.
In @.claude/agents/grant-hunter.md:
- Around line 52-56: The Focus Areas in .claude/agents/grant-hunter.md point to
disallowed top-level globs `docs/cost-ops/**` and `docs/cost-ops/programs/**`;
change those globs to an approved docs category (for example
`docs/operations/grants/**` and `docs/operations/programs/**` or another allowed
category per the repo guidelines) and update the Focus Areas section
accordingly, then regenerate the persona so the new paths are used everywhere.
In @.claude/commands/team-cost-ops.md:
- Line 85: The current "**Default handoff chain:** infra → product →
strategic-ops" is ambiguous about where cost-ops fits; update that heading (or
the surrounding sentence) to explicitly state cost-ops' role—either make it a
chain starting from cost-ops (e.g., "Default handoff chain: cost-ops →
strategic-ops → infra → product") or rename to "Default handoff targets" and
list that cost-ops hands off to infra → product → strategic-ops in priority
order; ensure you modify the line with the exact phrase "**Default handoff
chain:**" (or replace it with "**Default handoff targets:**") so readers know
whether cost-ops hands off to or receives from that chain.
In @.claude/hooks/protect-templates.sh:
- Around line 23-35: The path used to locate the repo package.json is incorrect:
change the AGENTKIT_PKG assignment (currently using SCRIPT_DIR and
../../../package.json) so it points to the repo root package.json (use
../../package.json from SCRIPT_DIR), keeping the existing existence check and
subsequent jq PKG_NAME read; update the AGENTKIT_PKG variable only (referenced
in the SCRIPT_DIR and PKG_NAME logic) so the agentkit-forge-runtime check can
correctly detect the source repo and exit early.
In @.github/agents/spec-compliance-auditor.agent.md:
- Around line 2-3: The YAML description scalar contains an unescaped apostrophe
in "agent's" which breaks single-quoted YAML; update the single-quoted
description value (the scalar following description:) by escaping the apostrophe
using YAML's doubling rule (change agent's to agent''s) so the parser can read
the string (refer to the description scalar in the block where name: 'Spec
Compliance Auditor' is defined).
In `@CHANGELOG.md`:
- Line 33: Update the CHANGELOG entry that currently references
"scripts/analyze-agents.mjs" to point to the actual wrapper scripts
"scripts/analyze-agents.sh" (Unix) and "scripts/analyze-agents.ps1" (Windows),
and if helpful mention the underlying engine at
".agentkit/engines/node/src/cli.mjs"; replace the `.mjs` reference with the
correct wrapper script names so the entry matches the real user-facing entry
points.
In `@docs/planning/agents-teams/strategic-ops-team.md`:
- Around line 37-40: This document and its sample scope use the disallowed
top-level docs/planning namespace; move the file out of docs/planning into the
appropriate approved category (for example docs/operations or docs/architecture
depending on content) and update the scope entry in the frontmatter: replace
'docs/planning/**' with the new category path (e.g., 'docs/operations/**' or
'docs/architecture/**') and ensure the scope array still contains the other
patterns ('.agentkit/spec/**', '**/*'); keep the id "strategic-ops" and name
"STRATEGIC OPS" unchanged.
In `@scripts/resolve-merge.sh`:
- Around line 1-4: This generated script should not be edited directly; update
the source spec files (.agentkit/spec/ or .agentkit/overlays/agentkit-forge) and
re-run pnpm -C .agentkit agentkit:sync to regenerate scripts, and while doing so
confirm the default branch used in the TARGET variable in
scripts/resolve-merge.sh (the TARGET="${1:-...}" assignment around where TARGET
is defined) matches your repo default—if your default branch is dev change the
default to TARGET="${1:-dev}" in the source spec before regenerating.
In `@scripts/setup-agentkit-branch-governance.ps1`:
- Around line 96-97: The two Write-Host lines both print the same verification
command (Write-Host "Verify with: gh api /repos/$Repo/branches/main/protection"
and the duplicate), so replace the second duplicated Write-Host line with the
intended branch verification (e.g., change "main" to the other branch variable
or name used in this script such as $Branch or $TargetBranch) so it prints the
correct gh api command for the second branch rather than repeating the main
branch check.
In `@scripts/setup-agentkit-branch-governance.sh`:
- Line 106: The for-loop `for BRANCH in main main` iterates over "main" twice
due to a copy-paste error; update the source spec under .agentkit/spec (or
overlay) to the correct branch list (e.g., `main` only or `dev main` as
intended) so the generated script no longer contains `for BRANCH in main main`,
then regenerate the script with `pnpm -C .agentkit agentkit:sync` so
scripts/setup-agentkit-branch-governance.sh is rebuilt with the corrected `for
BRANCH in ...` line.
- Around line 125-126: The two echo lines printing the same verification command
are duplicated; either remove the duplicate echo of "gh api
/repos/$REPO/branches/main/protection" or change them to show both branches you
protected (e.g., main and dev). Update the echo lines in
scripts/setup-agentkit-branch-governance.sh so they match the branches handled
by the script (use $REPO with /branches/main/protection and, if the script
protects dev as well, add /branches/dev/protection), ensuring the verification
hints reflect the actual branches modified rather than repeating the same line.
---
Outside diff comments:
In @.agentkit/templates/claude/hooks/protect-templates.sh:
- Around line 39-48: Normalize FILE_PATH before testing against
PROTECTED_PATTERNS so Windows backslashes can't bypass the checks: in
protect-templates.sh, before the for loop that iterates over PROTECTED_PATTERNS,
transform FILE_PATH to a normalized variable (e.g., replace backslashes with
forward slashes) and use that normalized path in the grep -qE check;
alternatively, update PROTECTED_PATTERNS to accept both '/' and '\' separators,
but ensure the script uses the normalized variable (refer to FILE_PATH,
PROTECTED_PATTERNS and the for pattern loop) when matching.
In @.claude/hooks/protect-templates.sh:
- Around line 42-53: The deny message references .agentkit/spec/ while that path
is still blocked in PROTECTED_PATTERNS, causing a loop; either remove the
'\.agentkit/spec/' entry from PROTECTED_PATTERNS to allow local spec edits, or
keep it blocked but update the jq --arg reason string used where FILE_PATH
matches to point to a writable configuration location (for example an overlays
or project config path) and include correct instructions (e.g., run 'pnpm -C
.agentkit agentkit:sync' only if that location is writable); modify the
PROTECTED_PATTERNS array or the --arg reason text accordingly so the guidance
does not point to a blocked path and reference PROTECTED_PATTERNS, FILE_PATH and
the jq --arg reason usage to locate the exact places to change.
In `@docs/planning/intake/intake-agent-proposal.md`:
- Around line 86-90: Update the Agent Definition's focus block so it uses the
canonical docs taxonomy instead of legacy paths: replace any occurrences of
"docs/design/**" with "docs/product/**" or "docs/architecture/**" as
appropriate, replace "docs/processes/**" with "docs/orchestration/**", and
replace "docs/prd/**" with "docs/product/**"; ensure these changes are made
within the Agent Definition's focus block so the intake agent will watch the
correct directories.
---
Duplicate comments:
In @.agentkit/spec/agents.yaml:
- Around line 725-731: The spec-compliance-auditor agent's focus list (the YAML
key "focus" under the spec for spec-compliance-auditor) is missing the output
pattern for its report directory; add "docs/agents/**" to the existing focus
array (alongside '.agentkit/spec/agents.yaml', '.agentkit/spec/teams.yaml',
'.claude/state/tasks/**', '.claude/state/events.log', 'docs/history/**') so the
agent can write its reports into the declared output location referenced by the
domain-rules.
In @.agentkit/templates/claude/hooks/protect-templates.sh:
- Around line 49-50: Update the denial message in protect-templates.sh so it
varies by repo type: in the jq --arg reason that references '${FILE_PATH}',
detect whether the current repo is the upstream agentkit-forge (e.g., check the
git remote URL or an env flag used in this script) and, if it is, keep the
existing guidance about opening a PR to agentkit-forge; otherwise (downstream
repos) replace the local-edit/sync instructions with a single instruction
telling users to propose changes by opening a PR against the agentkit-forge
repository. Locate the message construction around the jq --arg reason call in
protect-templates.sh and implement a simple if/else that sets the reason string
accordingly.
In @.agentkit/templates/scripts/resolve-merge.sh:
- Around line 79-89: The current loop uses grep -E on git diff output which
treats GENERATED_PATTERNS (e.g., `.agents/skills/*/SKILL.md`) as regexes and
breaks glob matching; update the loop that iterates over GENERATED_PATTERNS so
it filters conflicted files using shell glob matching instead of grep -E: read
filenames from git diff --name-only --diff-filter=U and for each file test the
glob against the pattern using a shell glob comparison (e.g., [[ "$file" ==
$pattern ]] or case) before running git checkout --theirs / git add; keep the
rest of the logic (ok message, auto_resolved increment) intact and remove the
grep -E usage.
In @.agentkit/templates/scripts/update-changelog.ps1:
- Around line 88-99: The insertion loop in update-changelog.ps1 (variables:
insertAt, sectionIdx, blockEnd, lines, entry) doesn't stop at a horizontal
divider, so add a check to break when the current line is the divider (e.g.,
lines[insertAt].trim() === '---' or lines[insertAt].startsWith('---')) alongside
the existing blank/heading checks; update the while condition to include
"!lines[insertAt].trim().startsWith('---')" (or equivalent) so the new entry is
inserted before the divider and inside the section.
In @.claude/hooks/protect-templates.ps1:
- Around line 29-39: The relative path used to detect the agentkit-forge source
repo is one level too high: update the $agentkitPkg assignment (which currently
joins $scriptDir with '..\..\..\package.json') to use one fewer parent segment
so it points at the repo root (e.g. '..\..\package.json' relative to
$scriptDir), then keep the existing Get-Content/ConvertFrom-Json check against
$pkg.name to allow the hook to exit when in the source repo.
---
Nitpick comments:
In @.agentkit/templates/scripts/analyze-agents.ps1:
- Around line 14-23: Add a preflight check that verifies the `node` executable
is available before invoking it: use a command check (e.g., Get-Command node or
equivalent) early in the script (before the node
"$ProjectRoot/.agentkit/engines/node/src/cli.mjs" call) and if not found write a
clear error message and exit with non-zero status; modify the script surrounding
the $ProjectRoot, $cliArgs and node invocation to bail out with a descriptive
message when `node` is missing so the subsequent node call is never attempted.
In @.agentkit/templates/scripts/create-doc.ps1:
- Around line 35-36: Replace positional arguments to Join-Path with named
parameters to improve readability: update the $HistoryDir assignment to call
Join-Path -Path $RepoRoot -ChildPath 'docs' and then Join-Path -Path (that
result) -ChildPath 'history' (or use nested Join-Path calls), and update
$IndexFile to call Join-Path -Path $HistoryDir -ChildPath '.index.json';
reference $HistoryDir, $IndexFile and Join-Path when making these changes.
In @.agentkit/templates/scripts/resolve-merge.ps1:
- Around line 70-89: Cache the result of the git conflicted-file query into a
single variable (currently $conflicted) once up front and reuse that array for
all subsequent operations instead of re-running `git diff --name-only
--diff-filter=U` multiple times; modify the code paths that currently re-run the
command to filter and operate on the in-memory $conflicted list (e.g., the loops
that use foreach ($file in $conflicted) and the checks against
$generatedPatterns) so all filtering (generated file pattern matching) and git
actions (git checkout --theirs, git add) operate on the cached list.
In @.agentkit/templates/scripts/setup-agentkit-branch-governance.ps1:
- Around line 77-81: The assignment to $branchExists is unused; update the gh
api call around the branch existence check so you either (A) remove the
$branchExists assignment and invoke gh api without capturing output (e.g.
discard stdout/stderr) and continue to check $LASTEXITCODE, or (B) capture the
API response into a meaningful variable (e.g. $response) and use that response
instead of relying solely on $LASTEXITCODE; modify the block referencing
$branchExists, $LASTEXITCODE, and the gh api invocation accordingly so no unused
variable remains.
In @.claude/commands/team-cost-ops.md:
- Line 282: Standardize the hyphenation of "follow-up" across the document by
replacing all instances of "followup" and "follow-ups" with the hyphenated form
"follow-up" (including in strings like the status name used in the orchestrator
such as "ready-for-followup" — update that to "ready-for-follow-up" if you
decide to hyphenate status names consistently) and any references in
handoffContext text; ensure consistency for both noun/adjective uses and for
status identifiers like "needs-review" vs "ready-for-follow-up" so the term is
uniform throughout.
In `@scripts/analyze-agents.ps1`:
- Line 26: Replace the inline string path to the Node CLI with a Join-Path-built
path and invoke it via the call operator so the script is platform-consistent:
build the CLI path from $ProjectRoot using Join-Path (targeting the
.agentkit/engines/node/src/cli.mjs segments) and then call node with that joined
path and the existing `@cliArgs`; reference $ProjectRoot, cli.mjs and `@cliArgs` to
locate where to change the invocation.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 513f3a5f-a30f-46a6-a584-67073fd380d0
📒 Files selected for processing (51)
.agentkit/engines/node/src/agent-analysis.mjs.agentkit/engines/node/src/cli.mjs.agentkit/spec/agents.yaml.agentkit/spec/rules.yaml.agentkit/spec/teams.yaml.agentkit/templates/claude/hooks/protect-templates.ps1.agentkit/templates/claude/hooks/protect-templates.sh.agentkit/templates/scripts/analyze-agents.ps1.agentkit/templates/scripts/create-doc.ps1.agentkit/templates/scripts/resolve-merge.ps1.agentkit/templates/scripts/resolve-merge.sh.agentkit/templates/scripts/setup-agentkit-branch-governance.ps1.agentkit/templates/scripts/setup-agentkit-branch-governance.sh.agentkit/templates/scripts/sync-split-pr.ps1.agentkit/templates/scripts/update-changelog.ps1.claude/agents/grant-hunter.md.claude/agents/spec-compliance-auditor.md.claude/agents/vendor-arbitrage-analyst.md.claude/commands/team-cost-ops.md.claude/commands/team-strategic-ops.md.claude/hooks/protect-templates.ps1.claude/hooks/protect-templates.sh.claude/rules/languages/ai-cost-ops.md.clinerules/ai-cost-ops.md.clinerules/languages/ai-cost-ops.md.cursor/rules/languages/ai-cost-ops.md.cursor/rules/team-strategic-ops.mdc.factory/settings.json.github/agents/grant-hunter.agent.md.github/agents/spec-compliance-auditor.agent.md.github/agents/vendor-arbitrage-analyst.agent.md.github/chatmodes/team-strategic-ops.chatmode.md.roo/rules/ai-cost-ops.md.roo/rules/languages/ai-cost-ops.md.windsurf/rules/languages/ai-cost-ops.md.windsurf/rules/team-strategic-ops.mdAGENT_BACKLOG.mdAGENT_TEAMS.mdCHANGELOG.mddocs/planning/agents-teams/forge-team.mddocs/planning/agents-teams/strategic-ops-team.mddocs/planning/cost-governance/multi-provider-infra-cost.mddocs/planning/intake/intake-agent-proposal.mdscripts/analyze-agents.ps1scripts/create-doc.ps1scripts/resolve-merge.ps1scripts/resolve-merge.shscripts/setup-agentkit-branch-governance.ps1scripts/setup-agentkit-branch-governance.shscripts/sync-split-pr.ps1scripts/update-changelog.ps1
🚧 Files skipped from review as they are similar to previous changes (11)
- AGENT_BACKLOG.md
- .github/chatmodes/team-strategic-ops.chatmode.md
- .agentkit/spec/teams.yaml
- .factory/settings.json
- .windsurf/rules/team-strategic-ops.md
- .clinerules/ai-cost-ops.md
- docs/planning/agents-teams/forge-team.md
- .claude/agents/vendor-arbitrage-analyst.md
- .agentkit/templates/scripts/setup-agentkit-branch-governance.sh
- .roo/rules/ai-cost-ops.md
- .clinerules/languages/ai-cost-ops.md
Collaborator
Author
CodeRabbit Review Response — Commit
|
Contributor
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In @.agentkit/templates/scripts/sync-split-pr.ps1:
- Around line 11-23: The script currently assumes native commands fail will
trigger PowerShell exceptions (e.g., gh auth status, pnpm, git
checkout/add/commit/push, gh pr create) but $ErrorActionPreference doesn't apply
to native executables; update the script to treat native command failures as
fatal by either setting $PSNativeCommandUseErrorActionPreference = $true at the
top or by adding a small helper (e.g., Invoke-NativeOrThrow) that runs a
scriptblock, checks $LASTEXITCODE and throws on non-zero exit; then wrap every
native call mentioned in the diff (gh auth status, pnpm -C .agentkit
agentkit:sync, git checkout, git add, git commit, git push, gh pr create) with
that mechanism so failures abort the script rather than continue.
In @.agentkit/templates/scripts/update-changelog.ps1:
- Around line 106-112: Do not modify the protected template
`.agentkit/templates/scripts/update-changelog.ps1`; instead update the
non-protected copy `scripts/update-changelog.ps1` to stop using
[System.IO.Path]::GetTempFileName() (which creates a real file) and instead
generate a temp pathname without creating the file—e.g. build $TmpScript via
[System.IO.Path]::Combine([System.IO.Path]::GetTempPath(),
[System.Guid]::NewGuid().ToString() + '.cjs')—then write to and later remove
$TmpScript; adjust the code that currently references $TmpScript and replace the
GetTempFileName usage to ensure no stray temp file remains.
- Around line 1-3: Replace the non-ASCII em dash in the console.log statement
with a plain ASCII hyphen: locate the console.log call in the script (the one
printing the changelog entry) and change the “—” character to "-" so the
generated .ps1 remains ASCII-only and PowerShell 5.1 won’t misdecode it; ensure
no other em dash characters remain in the file.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: d1a02bd8-20f4-42c1-8ede-5c7bdc33c944
📒 Files selected for processing (4)
.agentkit/templates/scripts/sync-split-pr.ps1.agentkit/templates/scripts/update-changelog.ps1scripts/sync-split-pr.ps1scripts/update-changelog.ps1
Linux CI sync produces +x permissions; align local to match. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
JustAGhosT
added a commit
that referenced
this pull request
Mar 11, 2026
* feat(teams): add Strategic Ops, Cost Ops teams and agent analysis engine (#356) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate all outputs after project review fixes Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(tests): isolate render target gating tests with fresh temp dirs (#382) Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
JustAGhosT
added a commit
that referenced
this pull request
Mar 11, 2026
* feat(teams): add Strategic Ops, Cost Ops teams and agent analysis engine (#356) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. * chore(sync): regenerate all outputs after project review fixes * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs --------- * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. --------- * fix(tests): isolate render target gating tests with fresh temp dirs (#382) Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
JustAGhosT
added a commit
that referenced
this pull request
Mar 11, 2026
…ility (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate all outputs after project review fixes Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate outputs after branch merges Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate outputs after merge of new-user-entry-point Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(engine): add brand color palette variables to sync vars Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
JustAGhosT
added a commit
that referenced
this pull request
Mar 12, 2026
* Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- Co-authored-by: Claude <noreply@anthropic.com> * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills - Add commandPrefix to vars from overlay settings - Add commandPrefixedName to buildCommandVars - Add 16 unit + integration tests (all pass, 93 existing tests unaffected) Default is null (no prefix) for full backwards compatibility. https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p * fix(sync): address review findings for command prefix - Remove unused afterAll import from test file - Add clarifying comment that non-spec command files are also prefixed - Add 2 integration tests verifying commandPrefixedName template variable renders correctly with and without prefix https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p --------- Co-authored-by: Claude <noreply@anthropic.com> * fix(ci): CI remediation — package manager, review findings, test stability (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate all outputs after project review fixes Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate outputs after branch merges Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate outputs after merge of new-user-entry-point Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(engine): add brand color palette variables to sync vars Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391) Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * Add entry point for new framework users (#389) * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * test(start): add coverage for result screen exit path Add App-level test that navigates through ConversationFlow to a leaf node and verifies the result screen text is rendered before exit(). Confirms the synchronous useEffect exit is safe — React commits the render (Ink captures the frame) before useEffect fires. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: update generated sync output timestamps Generated files updated with current sync date from dev merge. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW --------- Co-authored-by: Claude <noreply@anthropic.com> * fix: caldues heuristics (#398) --------- Co-authored-by: Claude <noreply@anthropic.com>
23 tasks
JustAGhosT
added a commit
that referenced
this pull request
Mar 12, 2026
* Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills - Add commandPrefix to vars from overlay settings - Add commandPrefixedName to buildCommandVars - Add 16 unit + integration tests (all pass, 93 existing tests unaffected) Default is null (no prefix) for full backwards compatibility. https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p * fix(sync): address review findings for command prefix - Remove unused afterAll import from test file - Add clarifying comment that non-spec command files are also prefixed - Add 2 integration tests verifying commandPrefixedName template variable renders correctly with and without prefix https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p --------- * fix(ci): CI remediation — package manager, review findings, test stability (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. * chore(sync): regenerate all outputs after project review fixes * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs --------- * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. * chore(sync): regenerate outputs after branch merges * chore(sync): regenerate outputs after merge of new-user-entry-point * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync * chore(engine): add brand color palette variables to sync vars * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. --------- * fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391) * Add entry point for new framework users (#389) * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * test(start): add coverage for result screen exit path Add App-level test that navigates through ConversationFlow to a leaf node and verifies the result screen text is rendered before exit(). Confirms the synchronous useEffect exit is safe — React commits the render (Ink captures the frame) before useEffect fires. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: update generated sync output timestamps Generated files updated with current sync date from dev merge. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW --------- * fix: caldues heuristics (#398) --------- Co-authored-by: Claude <noreply@anthropic.com>
JustAGhosT
added a commit
that referenced
this pull request
Mar 13, 2026
* Add start command: new user entry point with state detection (#387)
* fix(commands): add AskUserQuestion to VALID_TOOLS and /start command
The /start command session got stuck because AskUserQuestion was not
included in the allowed-tools whitelist. This fix addresses three issues:
1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite,
Agent, and NotebookEdit — preventing commands from declaring these
Claude Code built-in tools in their allowed-tools.
2. The /start command template now explicitly includes AskUserQuestion
in its allowed-tools frontmatter and instructs the agent to use it
for interactive guided choices (Phase 3).
3. Added /start command spec to commands.yaml with AskUserQuestion as
a declared tool dependency.
Also adds a test case validating all four newly-added tools are accepted
by the spec validator.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
* fix(commands): remove unrendered Handlebars comment from start.md output
The generated .claude/commands/start.md contained a raw {{! ... }}
Handlebars comment that was not processed by the sync engine. Remove it
so the generated output is clean Markdown.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
---------
Co-authored-by: Claude <noreply@anthropic.com>
* Add configurable prefix to kits commands (#388)
* feat(sync): add configurable command prefix for generated slash commands
Add `commandPrefix` setting to overlay settings that namespaces all
generated slash commands across platforms:
- Claude Code: subdirectory strategy (kits/check.md → /project:kits:check)
- Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check)
- Team commands excluded from prefixing (already namespaced)
Changes:
- Add resolveCommandPath() helper with subdirectory/filename strategies
- Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands,
syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills
- Add commandPrefix to vars from overlay settings
- Add commandPrefixedName to buildCommandVars
- Add 16 unit + integration tests (all pass, 93 existing tests unaffected)
Default is null (no prefix) for full backwards compatibility.
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
* fix(sync): address review findings for command prefix
- Remove unused afterAll import from test file
- Add clarifying comment that non-spec command files are also prefixed
- Add 2 integration tests verifying commandPrefixedName template variable
renders correctly with and without prefix
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(ci): CI remediation — package manager, review findings, test stability (#390)
* fix(ci): resolve 7 bugs from project review
- BUG-001: Replace flaky discover test with controlled temp fixture
- BUG-002: Run prettier --write to fix formatting drift
- BUG-003: Add form-template detection skip in issue label validation
- BUG-005: Change claude.yml to self-hosted runner
- BUG-006: Align branch protection required status checks with project.yaml
- BUG-007: Fix command injection in resolve-merge.sh (use grep -F)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: update changelog, add planning registry review findings
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed)
- Add Project Review Findings section to planning registry (PR-001 to PR-014)
- Update planning docs after sync merge
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(review): add --generate-plans flag to project-review command
Add Phase 2.5 plan generation after project review findings. When
--generate-plans is passed (default: true), scaffold plan files from
critical/high findings into docs/planning/review-findings/.
Also includes sync cleanup of stale cursor/windsurf settings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate all outputs after project review fixes
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cli): dynamic flag loading from commands.yaml + context-aware template hook
- Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags()
that reads flag definitions from commands.yaml at startup
- CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml
- Self-validation warns at startup if any flag is missing a type definition
- Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency
- Fix scaffold-once orphan bug: carry forward manifest entries for files
skipped by scaffold-once so orphan cleanup does not delete them
- Make protect-templates hook context-aware: skip protection in the
agentkit-forge source repo (detected via package.json name) so
maintainer agents can edit templates; block only in downstream repos
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(sync): add managed-mode script templates for downstream repos
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130)
Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating,
validating, and deploying new agent team specifications. Adapted from
cogmesh #130 with a simplified 6-agent pipeline:
- input-clarifier: assess requests, extract constraints
- mission-definer: lock team definition (ID, scope, accepts)
- role-architect: design agent roles and dependencies
- prompt-engineer: write agent descriptions and rules
- flow-designer: design team command and integration points
- team-validator: quality gate for spec consistency
Includes /team-forge command with --task flag (create-team, validate-team,
audit-teams, update-team) and planning doc.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(teams): add Strategic Ops team for cross-project coordination
Add the STRATEGIC OPS team (T12) — handles framework governance,
portfolio analysis, adoption strategy, impact assessment, and release
coordination across all repos using AgentKit Forge.
5-agent pipeline:
- portfolio-analyst: inventory repos, detect drift, adoption metrics
- governance-advisor: versioning strategy, breaking change protocols
- adoption-strategist: onboarding, migration paths, rollout plans
- impact-assessor: blast radius analysis for template/spec changes
- release-coordinator: version bumps, sync waves, release comms
Includes /team-strategic-ops command with --task and --scope flags.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(agents): add agent/team relationship matrix analysis engine + scripts
Add comprehensive agent/team relationship analysis with 8 cross-reference
matrices and 10 supplementary analyses (orphans, cycles, bottlenecks,
reachability, critical path, notification amplifiers, etc.).
- Fix YAML structure: strategic-ops agents now under own top-level key
- Add explicit agents: lists to forge + strategic-ops teams in teams.yaml
- Add consolidation detection responsibilities to portfolio-analyst
- Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers)
- Wire analyze-agents CLI command with --output/--matrix/--format flags
- Add managed-scaffold script templates (bash + PowerShell)
- Integrate into sync pipeline (auto-regenerates matrix on spec changes)
- Add 33 tests covering all matrices, analyses, and edge cases
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: update documentation files and add plan template
- Add trailing newlines to Cursor command documentation files for consistency
- Add new plan template files for project planning
- Improve markdown table formatting in Claude skills documentation
- Remove obsolete .clinerules/testing.md file
- Update various rule files with better formatting and advisory rule alignment
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364)
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365)
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): address CodeRabbit review findings on strategic-ops PR
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent
crash when node_modules is missing
- Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1
templates to fix invalid PowerShell syntax
- Fix protect-templates hook path: use 3 parent traversals to reach
.agentkit/package.json from hooks directory
- Remove schema-invalid 'phase' field from ai-cost-ops rule domain
- Narrow strategic-ops scope: replace **/* catch-all with specific files
- Add strategic-ops to cost-ops handoff-chain for consistency
- Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst,
and grant-hunter agents
- Fix glob matching in resolve-merge.sh (use regex instead of broken
sed strip)
- Add merge failure vs conflict detection in resolve-merge.ps1
- Add branch existence check in setup-agentkit-branch-governance scripts
- Add gh CLI preflight check in sync-split-pr.ps1
- Deduplicate branch protection loop when defaultBranch is 'main'
- Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/**
- Fix non-canonical doc paths in intake-agent-proposal.md
- Add changelog entries for new teams, agents, and analysis engine
- Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope)
- Run prettier on all modified files
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): add gh auth preflight and changelog divider handling
- sync-split-pr.ps1: add gh auth status check before side effects
- update-changelog.ps1: stop before --- divider when appending entries
Addresses CodeRabbit review comments #7 and #13 on PR #356.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(sync): set executable permission on analyze-agents.sh
Linux CI sync produces +x permissions; align local to match.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): address CodeRabbit review round 3-4 findings
- Fix YAML frontmatter in copilot agent template: use double quotes
for description field to handle apostrophes (CRITICAL)
- Fix protect-templates.sh/ps1 path traversal: correct parent
directory count for .agentkit/package.json source repo detection
- Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1
so native command failures (git, pnpm, gh) are treated as fatal
- Replace try/catch with $LASTEXITCODE check for gh auth status
- Deduplicate branch loop in setup-agentkit-branch-governance.sh
when defaultBranch equals 'main'
- Fix duplicate verification echo lines in governance scripts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* style(docs): fix prettier formatting on planning documents
Run prettier --write on web-intake-expansion.md and
intake-agent-proposal.md to fix CI formatting check.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tests): isolate render target gating tests with fresh temp dirs
Tests in the "render target gating" describe block shared a single temp
directory via beforeAll/afterAll. The first test ran --only claude
(generating .claude/ files), and the second ran --only warp expecting no
Claude files — but leftovers from test 1 caused the assertion to fail.
Changing to beforeEach/afterEach gives each test a clean directory.
Closes #377
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(engine): add pre-sync commit guard and interactive apply mode
Adds two safety features to the sync pipeline:
1. Pre-sync commit guard: Detects uncommitted changes in protected
directories (.agentkit/engines, spec, overlays) before sync runs.
In TTY mode, prompts to abort, stash, or continue. In non-TTY
mode (CI), prints a warning and proceeds.
2. Interactive apply mode: After rendering, shows a change summary
and prompts: apply all / skip all / prompt each file. Per-file
prompt supports show-diff and apply-all-remaining. Default in
TTY; bypassed with --yes, --no-prompt, or --force.
New module: sync-guard.mjs with 4 exported functions and 7 tests.
New CLI flags: --yes, --no-prompt for non-interactive sync.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate outputs after branch merges
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate outputs after merge of new-user-entry-point
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(engine): add configurable package manager and fix review findings
- Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn)
- Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.)
- Update CLAUDE.md template and hook/workflow templates to use {{packageManager}}
- Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch
- Remove duplicate cost-ops team definition from teams.yaml
- Pin all dependency versions in package.json (remove ^ prefixes)
- Add vitest coverage thresholds (80% statements/branches/functions/lines)
- Fix src/start/ code quality: null guards, exit delay, error boundaries
- Harden consolidate-branches.sh: self-resolution guard, stash restore
- Regenerate all 533 output files via agentkit sync
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(engine): add brand color palette variables to sync vars
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(ci): fix test race condition, workspace config, and lockfile
- Fix ConversationFlow test: wait for 'Got it' before asserting hint text
- Add packages field to pnpm-workspace.yaml for proper workspace resolution
- Regenerate lockfile after version pinning (removed ^ prefixes)
- Remove accidental .agentkit/templates/src/ artifacts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(start): add event loop yields to ConversationFlow tests
ink-select-input needs setImmediate yields after a new menu renders
before it can process ENTER keypresses. Without this, the second ENTER
in leaf-selection tests was swallowed, causing intermittent failures.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391)
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Add entry point for new framework users (#389)
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* test(start): add coverage for result screen exit path
Add App-level test that navigates through ConversationFlow to a leaf
node and verifies the result screen text is rendered before exit().
Confirms the synchronous useEffect exit is safe — React commits the
render (Ink captures the frame) before useEffect fires.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: update generated sync output timestamps
Generated files updated with current sync date from dev merge.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix: caldues heuristics (#398)
* feat: complete revisit of agents (#399) (#400)
* Add start command: new user entry point with state detection (#387)
* fix(commands): add AskUserQuestion to VALID_TOOLS and /start command
The /start command session got stuck because AskUserQuestion was not
included in the allowed-tools whitelist. This fix addresses three issues:
1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite,
Agent, and NotebookEdit — preventing commands from declaring these
Claude Code built-in tools in their allowed-tools.
2. The /start command template now explicitly includes AskUserQuestion
in its allowed-tools frontmatter and instructs the agent to use it
for interactive guided choices (Phase 3).
3. Added /start command spec to commands.yaml with AskUserQuestion as
a declared tool dependency.
Also adds a test case validating all four newly-added tools are accepted
by the spec validator.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
* fix(commands): remove unrendered Handlebars comment from start.md output
The generated .claude/commands/start.md contained a raw {{! ... }}
Handlebars comment that was not processed by the sync engine. Remove it
so the generated output is clean Markdown.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
---------
* Add configurable prefix to kits commands (#388)
* feat(sync): add configurable command prefix for generated slash commands
Add `commandPrefix` setting to overlay settings that namespaces all
generated slash commands across platforms:
- Claude Code: subdirectory strategy (kits/check.md → /project:kits:check)
- Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check)
- Team commands excluded from prefixing (already namespaced)
Changes:
- Add resolveCommandPath() helper with subdirectory/filename strategies
- Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands,
syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills
- Add commandPrefix to vars from overlay settings
- Add commandPrefixedName to buildCommandVars
- Add 16 unit + integration tests (all pass, 93 existing tests unaffected)
Default is null (no prefix) for full backwards compatibility.
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
* fix(sync): address review findings for command prefix
- Remove unused afterAll import from test file
- Add clarifying comment that non-spec command files are also prefixed
- Add 2 integration tests verifying commandPrefixedName template variable
renders correctly with and without prefix
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
---------
* fix(ci): CI remediation — package manager, review findings, test stability (#390)
* fix(ci): resolve 7 bugs from project review
- BUG-001: Replace flaky discover test with controlled temp fixture
- BUG-002: Run prettier --write to fix formatting drift
- BUG-003: Add form-template detection skip in issue label validation
- BUG-005: Change claude.yml to self-hosted runner
- BUG-006: Align branch protection required status checks with project.yaml
- BUG-007: Fix command injection in resolve-merge.sh (use grep -F)
* docs: update changelog, add planning registry review findings
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed)
- Add Project Review Findings section to planning registry (PR-001 to PR-014)
- Update planning docs after sync merge
* feat(review): add --generate-plans flag to project-review command
Add Phase 2.5 plan generation after project review findings. When
--generate-plans is passed (default: true), scaffold plan files from
critical/high findings into docs/planning/review-findings/.
Also includes sync cleanup of stale cursor/windsurf settings.
* chore(sync): regenerate all outputs after project review fixes
* feat(cli): dynamic flag loading from commands.yaml + context-aware template hook
- Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags()
that reads flag definitions from commands.yaml at startup
- CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml
- Self-validation warns at startup if any flag is missing a type definition
- Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency
- Fix scaffold-once orphan bug: carry forward manifest entries for files
skipped by scaffold-once so orphan cleanup does not delete them
- Make protect-templates hook context-aware: skip protection in the
agentkit-forge source repo (detected via package.json name) so
maintainer agents can edit templates; block only in downstream repos
* feat(sync): add managed-mode script templates for downstream repos
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
* feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130)
Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating,
validating, and deploying new agent team specifications. Adapted from
cogmesh #130 with a simplified 6-agent pipeline:
- input-clarifier: assess requests, extract constraints
- mission-definer: lock team definition (ID, scope, accepts)
- role-architect: design agent roles and dependencies
- prompt-engineer: write agent descriptions and rules
- flow-designer: design team command and integration points
- team-validator: quality gate for spec consistency
Includes /team-forge command with --task flag (create-team, validate-team,
audit-teams, update-team) and planning doc.
* feat(teams): add Strategic Ops team for cross-project coordination
Add the STRATEGIC OPS team (T12) — handles framework governance,
portfolio analysis, adoption strategy, impact assessment, and release
coordination across all repos using AgentKit Forge.
5-agent pipeline:
- portfolio-analyst: inventory repos, detect drift, adoption metrics
- governance-advisor: versioning strategy, breaking change protocols
- adoption-strategist: onboarding, migration paths, rollout plans
- impact-assessor: blast radius analysis for template/spec changes
- release-coordinator: version bumps, sync waves, release comms
Includes /team-strategic-ops command with --task and --scope flags.
* feat(agents): add agent/team relationship matrix analysis engine + scripts
Add comprehensive agent/team relationship analysis with 8 cross-reference
matrices and 10 supplementary analyses (orphans, cycles, bottlenecks,
reachability, critical path, notification amplifiers, etc.).
- Fix YAML structure: strategic-ops agents now under own top-level key
- Add explicit agents: lists to forge + strategic-ops teams in teams.yaml
- Add consolidation detection responsibilities to portfolio-analyst
- Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers)
- Wire analyze-agents CLI command with --output/--matrix/--format flags
- Add managed-scaffold script templates (bash + PowerShell)
- Integrate into sync pipeline (auto-regenerates matrix on spec changes)
- Add 33 tests covering all matrices, analyses, and edge cases
* chore: update documentation files and add plan template
- Add trailing newlines to Cursor command documentation files for consistency
- Add new plan template files for project planning
- Improve markdown table formatting in Claude skills documentation
- Remove obsolete .clinerules/testing.md file
- Update various rule files with better formatting and advisory rule alignment
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
* feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364)
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365)
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
---------
* fix(teams): address CodeRabbit review findings on strategic-ops PR
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent
crash when node_modules is missing
- Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1
templates to fix invalid PowerShell syntax
- Fix protect-templates hook path: use 3 parent traversals to reach
.agentkit/package.json from hooks directory
- Remove schema-invalid 'phase' field from ai-cost-ops rule domain
- Narrow strategic-ops scope: replace **/* catch-all with specific files
- Add strategic-ops to cost-ops handoff-chain for consistency
- Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst,
and grant-hunter agents
- Fix glob matching in resolve-merge.sh (use regex instead of broken
sed strip)
- Add merge failure vs conflict detection in resolve-merge.ps1
- Add branch existence check in setup-agentkit-branch-governance scripts
- Add gh CLI preflight check in sync-split-pr.ps1
- Deduplicate branch protection loop when defaultBranch is 'main'
- Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/**
- Fix non-canonical doc paths in intake-agent-proposal.md
- Add changelog entries for new teams, agents, and analysis engine
- Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope)
- Run prettier on all modified files
* fix(templates): add gh auth preflight and changelog divider handling
- sync-split-pr.ps1: add gh auth status check before side effects
- update-changelog.ps1: stop before --- divider when appending entries
Addresses CodeRabbit review comments #7 and #13 on PR #356.
* fix(sync): set executable permission on analyze-agents.sh
Linux CI sync produces +x permissions; align local to match.
* fix(templates): address CodeRabbit review round 3-4 findings
- Fix YAML frontmatter in copilot agent template: use double quotes
for description field to handle apostrophes (CRITICAL)
- Fix protect-templates.sh/ps1 path traversal: correct parent
directory count for .agentkit/package.json source repo detection
- Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1
so native command failures (git, pnpm, gh) are treated as fatal
- Replace try/catch with $LASTEXITCODE check for gh auth status
- Deduplicate branch loop in setup-agentkit-branch-governance.sh
when defaultBranch equals 'main'
- Fix duplicate verification echo lines in governance scripts
* style(docs): fix prettier formatting on planning documents
Run prettier --write on web-intake-expansion.md and
intake-agent-proposal.md to fix CI formatting check.
* fix(tests): isolate render target gating tests with fresh temp dirs
Tests in the "render target gating" describe block shared a single temp
directory via beforeAll/afterAll. The first test ran --only claude
(generating .claude/ files), and the second ran --only warp expecting no
Claude files — but leftovers from test 1 caused the assertion to fail.
Changing to beforeEach/afterEach gives each test a clean directory.
Closes #377
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(engine): add pre-sync commit guard and interactive apply mode
Adds two safety features to the sync pipeline:
1. Pre-sync commit guard: Detects uncommitted changes in protected
directories (.agentkit/engines, spec, overlays) before sync runs.
In TTY mode, prompts to abort, stash, or continue. In non-TTY
mode (CI), prints a warning and proceeds.
2. Interactive apply mode: After rendering, shows a change summary
and prompts: apply all / skip all / prompt each file. Per-file
prompt supports show-diff and apply-all-remaining. Default in
TTY; bypassed with --yes, --no-prompt, or --force.
New module: sync-guard.mjs with 4 exported functions and 7 tests.
New CLI flags: --yes, --no-prompt for non-interactive sync.
* chore(sync): regenerate outputs after branch merges
* chore(sync): regenerate outputs after merge of new-user-entry-point
* feat(engine): add configurable package manager and fix review findings
- Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn)
- Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.)
- Update CLAUDE.md template and hook/workflow templates to use {{packageManager}}
- Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch
- Remove duplicate cost-ops team definition from teams.yaml
- Pin all dependency versions in package.json (remove ^ prefixes)
- Add vitest coverage thresholds (80% statements/branches/functions/lines)
- Fix src/start/ code quality: null guards, exit delay, error boundaries
- Harden consolidate-branches.sh: self-resolution guard, stash restore
- Regenerate all 533 output files via agentkit sync
* chore(engine): add brand color palette variables to sync vars
* fix(ci): fix test race condition, workspace config, and lockfile
- Fix ConversationFlow test: wait for 'Got it' before asserting hint text
- Add packages field to pnpm-workspace.yaml for proper workspace resolution
- Regenerate lockfile after version pinning (removed ^ prefixes)
- Remove accidental .agentkit/templates/src/ artifacts
* fix(start): add event loop yields to ConversationFlow tests
ink-select-input needs setImmediate yields after a new menu renders
before it can process ENTER keypresses. Without this, the second ENTER
in leaf-selection tests was swallowed, causing intermittent failures.
---------
* fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391)
* Add entry point for new framework users (#389)
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* test(start): add coverage for result screen exit path
Add App-level test that navigates through ConversationFlow to a leaf
node and verifies the result screen text is rendered before exit().
Confirms the synchronous useEffect exit is safe — React commits the
render (Ink captures the frame) before useEffect fires.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: update generated sync output timestamps
Generated files updated with current sync date from dev merge.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
---------
* fix: caldues heuristics (#398)
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(quality): resolve all lint and format errors
- Fix Prettier formatting across engine src and start components
- Resolve 381 markdown lint errors (MD040, MD024, MD036, MD029, MD022, MD001)
- Add markdownlint config rules for intentional doc patterns
- Add .claude/worktrees to prettierignore to exclude external branches
- Enable markdownlint MD024/MD026 for duplicate headings and trailing colons
All quality checks now pass: Prettier (0 errors), Markdown lint (0 errors), Tests (111/111), Build (28.9kb)
* chore(sync): regenerate AI tool configs from spec
- Run agentkit sync to update all generated configs
- Update skill files across all render targets
- 533 files regenerated
---------
Co-authored-by: Claude <noreply@anthropic.com>
JustAGhosT
added a commit
that referenced
this pull request
Mar 15, 2026
) * Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- Co-authored-by: Claude <noreply@anthropic.com> * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills - Add commandPrefix to vars from overlay settings - Add commandPrefixedName to buildCommandVars - Add 16 unit + integration tests (all pass, 93 existing tests unaffected) Default is null (no prefix) for full backwards compatibility. https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p * fix(sync): address review findings for command prefix - Remove unused afterAll import from test file - Add clarifying comment that non-spec command files are also prefixed - Add 2 integration tests verifying commandPrefixedName template variable renders correctly with and without prefix https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p --------- Co-authored-by: Claude <noreply@anthropic.com> * fix(ci): CI remediation — package manager, review findings, test stability (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate all outputs after project review fixes Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate outputs after branch merges Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate outputs after merge of new-user-entry-point Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(engine): add brand color palette variables to sync vars Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391) Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * Add entry point for new framework users (#389) * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * test(start): add coverage for result screen exit path Add App-level test that navigates through ConversationFlow to a leaf node and verifies the result screen text is rendered before exit(). Confirms the synchronous useEffect exit is safe — React commits the render (Ink captures the frame) before useEffect fires. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: update generated sync output timestamps Generated files updated with current sync date from dev merge. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW --------- Co-authored-by: Claude <noreply@anthropic.com> * fix: caldues heuristics (#398) * feat: complete revisit of agents (#399) (#400) * Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills - Add commandPrefix to vars from overlay settings - Add commandPrefixedName to buildCommandVars - Add 16 unit + integration tests (all pass, 93 existing tests unaffected) Default is null (no prefix) for full backwards compatibility. https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p * fix(sync): address review findings for command prefix - Remove unused afterAll import from test file - Add clarifying comment that non-spec command files are also prefixed - Add 2 integration tests verifying commandPrefixedName template variable renders correctly with and without prefix https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p --------- * fix(ci): CI remediation — package manager, review findings, test stability (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. * chore(sync): regenerate all outputs after project review fixes * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs --------- * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. * chore(sync): regenerate outputs after branch merges * chore(sync): regenerate outputs after merge of new-user-entry-point * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync * chore(engine): add brand color palette variables to sync vars * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. --------- * fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391) * Add entry point for new framework users (#389) * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * test(start): add coverage for result screen exit path Add App-level test that navigates through ConversationFlow to a leaf node and verifies the result screen text is rendered before exit(). Confirms the synchronous useEffect exit is safe — React commits the render (Ink captures the frame) before useEffect fires. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: update generated sync output timestamps Generated files updated with current sync date from dev merge. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW --------- * fix: caldues heuristics (#398) --------- Co-authored-by: Claude <noreply@anthropic.com> * fix(quality): resolve all lint and format errors - Fix Prettier formatting across engine src and start components - Resolve 381 markdown lint errors (MD040, MD024, MD036, MD029, MD022, MD001) - Add markdownlint config rules for intentional doc patterns - Add .claude/worktrees to prettierignore to exclude external branches - Enable markdownlint MD024/MD026 for duplicate headings and trailing colons All quality checks now pass: Prettier (0 errors), Markdown lint (0 errors), Tests (111/111), Build (28.9kb) * I'll update the last_updated field in all the files from '2026-03-12' to '2026-03-13'. This appears to be a routine timestamp update across all the AgentKit Forge generated files. * docs: add AgentKit Forge sync feedback Add detailed feedback on AgentKit Forge v3.1.0 integration, focusing on: - Windows line-ending issues and multi-editor sync behavior - Documentation scaffold-once limitations and override challenges - Unresolved placeholder warnings lacking diagnostics - Windows-specific pnpm execution problems * docs: update CLAUDE.md with repository-specific editing guidelines Clarified that modifications to `.agentkit` files are permitted in the agentkit-forge repository, while upstream directories remain protected. This ensures users understand the editing boundaries for project configuration and template files. * docs(claude): allow .agentkit edits in this repo (NB for framework dev) Made-with: Cursor * chore(sync): regenerate outputs after agentkit:sync Made-with: Cursor * docs: update AGENT_BACKLOG.md with detailed task scopes Enhanced task descriptions for CI pipeline configuration and test framework setup, specifying additional scopes for branch protection, drift checks, and quality gates on the main branch, as well as coverage targets for the .agentkit test suite. --------- Co-authored-by: Claude <noreply@anthropic.com>
JustAGhosT
added a commit
that referenced
this pull request
Mar 15, 2026
* Add start command: new user entry point with state detection (#387)
* fix(commands): add AskUserQuestion to VALID_TOOLS and /start command
The /start command session got stuck because AskUserQuestion was not
included in the allowed-tools whitelist. This fix addresses three issues:
1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite,
Agent, and NotebookEdit — preventing commands from declaring these
Claude Code built-in tools in their allowed-tools.
2. The /start command template now explicitly includes AskUserQuestion
in its allowed-tools frontmatter and instructs the agent to use it
for interactive guided choices (Phase 3).
3. Added /start command spec to commands.yaml with AskUserQuestion as
a declared tool dependency.
Also adds a test case validating all four newly-added tools are accepted
by the spec validator.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
* fix(commands): remove unrendered Handlebars comment from start.md output
The generated .claude/commands/start.md contained a raw {{! ... }}
Handlebars comment that was not processed by the sync engine. Remove it
so the generated output is clean Markdown.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
---------
Co-authored-by: Claude <noreply@anthropic.com>
* Add configurable prefix to kits commands (#388)
* feat(sync): add configurable command prefix for generated slash commands
Add `commandPrefix` setting to overlay settings that namespaces all
generated slash commands across platforms:
- Claude Code: subdirectory strategy (kits/check.md → /project:kits:check)
- Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check)
- Team commands excluded from prefixing (already namespaced)
Changes:
- Add resolveCommandPath() helper with subdirectory/filename strategies
- Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands,
syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills
- Add commandPrefix to vars from overlay settings
- Add commandPrefixedName to buildCommandVars
- Add 16 unit + integration tests (all pass, 93 existing tests unaffected)
Default is null (no prefix) for full backwards compatibility.
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
* fix(sync): address review findings for command prefix
- Remove unused afterAll import from test file
- Add clarifying comment that non-spec command files are also prefixed
- Add 2 integration tests verifying commandPrefixedName template variable
renders correctly with and without prefix
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(ci): CI remediation — package manager, review findings, test stability (#390)
* fix(ci): resolve 7 bugs from project review
- BUG-001: Replace flaky discover test with controlled temp fixture
- BUG-002: Run prettier --write to fix formatting drift
- BUG-003: Add form-template detection skip in issue label validation
- BUG-005: Change claude.yml to self-hosted runner
- BUG-006: Align branch protection required status checks with project.yaml
- BUG-007: Fix command injection in resolve-merge.sh (use grep -F)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: update changelog, add planning registry review findings
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed)
- Add Project Review Findings section to planning registry (PR-001 to PR-014)
- Update planning docs after sync merge
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(review): add --generate-plans flag to project-review command
Add Phase 2.5 plan generation after project review findings. When
--generate-plans is passed (default: true), scaffold plan files from
critical/high findings into docs/planning/review-findings/.
Also includes sync cleanup of stale cursor/windsurf settings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate all outputs after project review fixes
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cli): dynamic flag loading from commands.yaml + context-aware template hook
- Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags()
that reads flag definitions from commands.yaml at startup
- CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml
- Self-validation warns at startup if any flag is missing a type definition
- Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency
- Fix scaffold-once orphan bug: carry forward manifest entries for files
skipped by scaffold-once so orphan cleanup does not delete them
- Make protect-templates hook context-aware: skip protection in the
agentkit-forge source repo (detected via package.json name) so
maintainer agents can edit templates; block only in downstream repos
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(sync): add managed-mode script templates for downstream repos
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130)
Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating,
validating, and deploying new agent team specifications. Adapted from
cogmesh #130 with a simplified 6-agent pipeline:
- input-clarifier: assess requests, extract constraints
- mission-definer: lock team definition (ID, scope, accepts)
- role-architect: design agent roles and dependencies
- prompt-engineer: write agent descriptions and rules
- flow-designer: design team command and integration points
- team-validator: quality gate for spec consistency
Includes /team-forge command with --task flag (create-team, validate-team,
audit-teams, update-team) and planning doc.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(teams): add Strategic Ops team for cross-project coordination
Add the STRATEGIC OPS team (T12) — handles framework governance,
portfolio analysis, adoption strategy, impact assessment, and release
coordination across all repos using AgentKit Forge.
5-agent pipeline:
- portfolio-analyst: inventory repos, detect drift, adoption metrics
- governance-advisor: versioning strategy, breaking change protocols
- adoption-strategist: onboarding, migration paths, rollout plans
- impact-assessor: blast radius analysis for template/spec changes
- release-coordinator: version bumps, sync waves, release comms
Includes /team-strategic-ops command with --task and --scope flags.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(agents): add agent/team relationship matrix analysis engine + scripts
Add comprehensive agent/team relationship analysis with 8 cross-reference
matrices and 10 supplementary analyses (orphans, cycles, bottlenecks,
reachability, critical path, notification amplifiers, etc.).
- Fix YAML structure: strategic-ops agents now under own top-level key
- Add explicit agents: lists to forge + strategic-ops teams in teams.yaml
- Add consolidation detection responsibilities to portfolio-analyst
- Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers)
- Wire analyze-agents CLI command with --output/--matrix/--format flags
- Add managed-scaffold script templates (bash + PowerShell)
- Integrate into sync pipeline (auto-regenerates matrix on spec changes)
- Add 33 tests covering all matrices, analyses, and edge cases
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: update documentation files and add plan template
- Add trailing newlines to Cursor command documentation files for consistency
- Add new plan template files for project planning
- Improve markdown table formatting in Claude skills documentation
- Remove obsolete .clinerules/testing.md file
- Update various rule files with better formatting and advisory rule alignment
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364)
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365)
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): address CodeRabbit review findings on strategic-ops PR
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent
crash when node_modules is missing
- Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1
templates to fix invalid PowerShell syntax
- Fix protect-templates hook path: use 3 parent traversals to reach
.agentkit/package.json from hooks directory
- Remove schema-invalid 'phase' field from ai-cost-ops rule domain
- Narrow strategic-ops scope: replace **/* catch-all with specific files
- Add strategic-ops to cost-ops handoff-chain for consistency
- Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst,
and grant-hunter agents
- Fix glob matching in resolve-merge.sh (use regex instead of broken
sed strip)
- Add merge failure vs conflict detection in resolve-merge.ps1
- Add branch existence check in setup-agentkit-branch-governance scripts
- Add gh CLI preflight check in sync-split-pr.ps1
- Deduplicate branch protection loop when defaultBranch is 'main'
- Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/**
- Fix non-canonical doc paths in intake-agent-proposal.md
- Add changelog entries for new teams, agents, and analysis engine
- Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope)
- Run prettier on all modified files
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): add gh auth preflight and changelog divider handling
- sync-split-pr.ps1: add gh auth status check before side effects
- update-changelog.ps1: stop before --- divider when appending entries
Addresses CodeRabbit review comments #7 and #13 on PR #356.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(sync): set executable permission on analyze-agents.sh
Linux CI sync produces +x permissions; align local to match.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): address CodeRabbit review round 3-4 findings
- Fix YAML frontmatter in copilot agent template: use double quotes
for description field to handle apostrophes (CRITICAL)
- Fix protect-templates.sh/ps1 path traversal: correct parent
directory count for .agentkit/package.json source repo detection
- Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1
so native command failures (git, pnpm, gh) are treated as fatal
- Replace try/catch with $LASTEXITCODE check for gh auth status
- Deduplicate branch loop in setup-agentkit-branch-governance.sh
when defaultBranch equals 'main'
- Fix duplicate verification echo lines in governance scripts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* style(docs): fix prettier formatting on planning documents
Run prettier --write on web-intake-expansion.md and
intake-agent-proposal.md to fix CI formatting check.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tests): isolate render target gating tests with fresh temp dirs
Tests in the "render target gating" describe block shared a single temp
directory via beforeAll/afterAll. The first test ran --only claude
(generating .claude/ files), and the second ran --only warp expecting no
Claude files — but leftovers from test 1 caused the assertion to fail.
Changing to beforeEach/afterEach gives each test a clean directory.
Closes #377
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(engine): add pre-sync commit guard and interactive apply mode
Adds two safety features to the sync pipeline:
1. Pre-sync commit guard: Detects uncommitted changes in protected
directories (.agentkit/engines, spec, overlays) before sync runs.
In TTY mode, prompts to abort, stash, or continue. In non-TTY
mode (CI), prints a warning and proceeds.
2. Interactive apply mode: After rendering, shows a change summary
and prompts: apply all / skip all / prompt each file. Per-file
prompt supports show-diff and apply-all-remaining. Default in
TTY; bypassed with --yes, --no-prompt, or --force.
New module: sync-guard.mjs with 4 exported functions and 7 tests.
New CLI flags: --yes, --no-prompt for non-interactive sync.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate outputs after branch merges
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate outputs after merge of new-user-entry-point
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(engine): add configurable package manager and fix review findings
- Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn)
- Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.)
- Update CLAUDE.md template and hook/workflow templates to use {{packageManager}}
- Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch
- Remove duplicate cost-ops team definition from teams.yaml
- Pin all dependency versions in package.json (remove ^ prefixes)
- Add vitest coverage thresholds (80% statements/branches/functions/lines)
- Fix src/start/ code quality: null guards, exit delay, error boundaries
- Harden consolidate-branches.sh: self-resolution guard, stash restore
- Regenerate all 533 output files via agentkit sync
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(engine): add brand color palette variables to sync vars
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(ci): fix test race condition, workspace config, and lockfile
- Fix ConversationFlow test: wait for 'Got it' before asserting hint text
- Add packages field to pnpm-workspace.yaml for proper workspace resolution
- Regenerate lockfile after version pinning (removed ^ prefixes)
- Remove accidental .agentkit/templates/src/ artifacts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(start): add event loop yields to ConversationFlow tests
ink-select-input needs setImmediate yields after a new menu renders
before it can process ENTER keypresses. Without this, the second ENTER
in leaf-selection tests was swallowed, causing intermittent failures.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391)
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Add entry point for new framework users (#389)
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* test(start): add coverage for result screen exit path
Add App-level test that navigates through ConversationFlow to a leaf
node and verifies the result screen text is rendered before exit().
Confirms the synchronous useEffect exit is safe — React commits the
render (Ink captures the frame) before useEffect fires.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: update generated sync output timestamps
Generated files updated with current sync date from dev merge.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix: caldues heuristics (#398)
* feat: complete revisit of agents (#399) (#400)
* Add start command: new user entry point with state detection (#387)
* fix(commands): add AskUserQuestion to VALID_TOOLS and /start command
The /start command session got stuck because AskUserQuestion was not
included in the allowed-tools whitelist. This fix addresses three issues:
1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite,
Agent, and NotebookEdit — preventing commands from declaring these
Claude Code built-in tools in their allowed-tools.
2. The /start command template now explicitly includes AskUserQuestion
in its allowed-tools frontmatter and instructs the agent to use it
for interactive guided choices (Phase 3).
3. Added /start command spec to commands.yaml with AskUserQuestion as
a declared tool dependency.
Also adds a test case validating all four newly-added tools are accepted
by the spec validator.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
* fix(commands): remove unrendered Handlebars comment from start.md output
The generated .claude/commands/start.md contained a raw {{! ... }}
Handlebars comment that was not processed by the sync engine. Remove it
so the generated output is clean Markdown.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
---------
* Add configurable prefix to kits commands (#388)
* feat(sync): add configurable command prefix for generated slash commands
Add `commandPrefix` setting to overlay settings that namespaces all
generated slash commands across platforms:
- Claude Code: subdirectory strategy (kits/check.md → /project:kits:check)
- Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check)
- Team commands excluded from prefixing (already namespaced)
Changes:
- Add resolveCommandPath() helper with subdirectory/filename strategies
- Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands,
syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills
- Add commandPrefix to vars from overlay settings
- Add commandPrefixedName to buildCommandVars
- Add 16 unit + integration tests (all pass, 93 existing tests unaffected)
Default is null (no prefix) for full backwards compatibility.
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
* fix(sync): address review findings for command prefix
- Remove unused afterAll import from test file
- Add clarifying comment that non-spec command files are also prefixed
- Add 2 integration tests verifying commandPrefixedName template variable
renders correctly with and without prefix
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
---------
* fix(ci): CI remediation — package manager, review findings, test stability (#390)
* fix(ci): resolve 7 bugs from project review
- BUG-001: Replace flaky discover test with controlled temp fixture
- BUG-002: Run prettier --write to fix formatting drift
- BUG-003: Add form-template detection skip in issue label validation
- BUG-005: Change claude.yml to self-hosted runner
- BUG-006: Align branch protection required status checks with project.yaml
- BUG-007: Fix command injection in resolve-merge.sh (use grep -F)
* docs: update changelog, add planning registry review findings
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed)
- Add Project Review Findings section to planning registry (PR-001 to PR-014)
- Update planning docs after sync merge
* feat(review): add --generate-plans flag to project-review command
Add Phase 2.5 plan generation after project review findings. When
--generate-plans is passed (default: true), scaffold plan files from
critical/high findings into docs/planning/review-findings/.
Also includes sync cleanup of stale cursor/windsurf settings.
* chore(sync): regenerate all outputs after project review fixes
* feat(cli): dynamic flag loading from commands.yaml + context-aware template hook
- Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags()
that reads flag definitions from commands.yaml at startup
- CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml
- Self-validation warns at startup if any flag is missing a type definition
- Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency
- Fix scaffold-once orphan bug: carry forward manifest entries for files
skipped by scaffold-once so orphan cleanup does not delete them
- Make protect-templates hook context-aware: skip protection in the
agentkit-forge source repo (detected via package.json name) so
maintainer agents can edit templates; block only in downstream repos
* feat(sync): add managed-mode script templates for downstream repos
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
* feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130)
Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating,
validating, and deploying new agent team specifications. Adapted from
cogmesh #130 with a simplified 6-agent pipeline:
- input-clarifier: assess requests, extract constraints
- mission-definer: lock team definition (ID, scope, accepts)
- role-architect: design agent roles and dependencies
- prompt-engineer: write agent descriptions and rules
- flow-designer: design team command and integration points
- team-validator: quality gate for spec consistency
Includes /team-forge command with --task flag (create-team, validate-team,
audit-teams, update-team) and planning doc.
* feat(teams): add Strategic Ops team for cross-project coordination
Add the STRATEGIC OPS team (T12) — handles framework governance,
portfolio analysis, adoption strategy, impact assessment, and release
coordination across all repos using AgentKit Forge.
5-agent pipeline:
- portfolio-analyst: inventory repos, detect drift, adoption metrics
- governance-advisor: versioning strategy, breaking change protocols
- adoption-strategist: onboarding, migration paths, rollout plans
- impact-assessor: blast radius analysis for template/spec changes
- release-coordinator: version bumps, sync waves, release comms
Includes /team-strategic-ops command with --task and --scope flags.
* feat(agents): add agent/team relationship matrix analysis engine + scripts
Add comprehensive agent/team relationship analysis with 8 cross-reference
matrices and 10 supplementary analyses (orphans, cycles, bottlenecks,
reachability, critical path, notification amplifiers, etc.).
- Fix YAML structure: strategic-ops agents now under own top-level key
- Add explicit agents: lists to forge + strategic-ops teams in teams.yaml
- Add consolidation detection responsibilities to portfolio-analyst
- Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers)
- Wire analyze-agents CLI command with --output/--matrix/--format flags
- Add managed-scaffold script templates (bash + PowerShell)
- Integrate into sync pipeline (auto-regenerates matrix on spec changes)
- Add 33 tests covering all matrices, analyses, and edge cases
* chore: update documentation files and add plan template
- Add trailing newlines to Cursor command documentation files for consistency
- Add new plan template files for project planning
- Improve markdown table formatting in Claude skills documentation
- Remove obsolete .clinerules/testing.md file
- Update various rule files with better formatting and advisory rule alignment
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
* feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364)
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365)
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
---------
* fix(teams): address CodeRabbit review findings on strategic-ops PR
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent
crash when node_modules is missing
- Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1
templates to fix invalid PowerShell syntax
- Fix protect-templates hook path: use 3 parent traversals to reach
.agentkit/package.json from hooks directory
- Remove schema-invalid 'phase' field from ai-cost-ops rule domain
- Narrow strategic-ops scope: replace **/* catch-all with specific files
- Add strategic-ops to cost-ops handoff-chain for consistency
- Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst,
and grant-hunter agents
- Fix glob matching in resolve-merge.sh (use regex instead of broken
sed strip)
- Add merge failure vs conflict detection in resolve-merge.ps1
- Add branch existence check in setup-agentkit-branch-governance scripts
- Add gh CLI preflight check in sync-split-pr.ps1
- Deduplicate branch protection loop when defaultBranch is 'main'
- Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/**
- Fix non-canonical doc paths in intake-agent-proposal.md
- Add changelog entries for new teams, agents, and analysis engine
- Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope)
- Run prettier on all modified files
* fix(templates): add gh auth preflight and changelog divider handling
- sync-split-pr.ps1: add gh auth status check before side effects
- update-changelog.ps1: stop before --- divider when appending entries
Addresses CodeRabbit review comments #7 and #13 on PR #356.
* fix(sync): set executable permission on analyze-agents.sh
Linux CI sync produces +x permissions; align local to match.
* fix(templates): address CodeRabbit review round 3-4 findings
- Fix YAML frontmatter in copilot agent template: use double quotes
for description field to handle apostrophes (CRITICAL)
- Fix protect-templates.sh/ps1 path traversal: correct parent
directory count for .agentkit/package.json source repo detection
- Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1
so native command failures (git, pnpm, gh) are treated as fatal
- Replace try/catch with $LASTEXITCODE check for gh auth status
- Deduplicate branch loop in setup-agentkit-branch-governance.sh
when defaultBranch equals 'main'
- Fix duplicate verification echo lines in governance scripts
* style(docs): fix prettier formatting on planning documents
Run prettier --write on web-intake-expansion.md and
intake-agent-proposal.md to fix CI formatting check.
* fix(tests): isolate render target gating tests with fresh temp dirs
Tests in the "render target gating" describe block shared a single temp
directory via beforeAll/afterAll. The first test ran --only claude
(generating .claude/ files), and the second ran --only warp expecting no
Claude files — but leftovers from test 1 caused the assertion to fail.
Changing to beforeEach/afterEach gives each test a clean directory.
Closes #377
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(engine): add pre-sync commit guard and interactive apply mode
Adds two safety features to the sync pipeline:
1. Pre-sync commit guard: Detects uncommitted changes in protected
directories (.agentkit/engines, spec, overlays) before sync runs.
In TTY mode, prompts to abort, stash, or continue. In non-TTY
mode (CI), prints a warning and proceeds.
2. Interactive apply mode: After rendering, shows a change summary
and prompts: apply all / skip all / prompt each file. Per-file
prompt supports show-diff and apply-all-remaining. Default in
TTY; bypassed with --yes, --no-prompt, or --force.
New module: sync-guard.mjs with 4 exported functions and 7 tests.
New CLI flags: --yes, --no-prompt for non-interactive sync.
* chore(sync): regenerate outputs after branch merges
* chore(sync): regenerate outputs after merge of new-user-entry-point
* feat(engine): add configurable package manager and fix review findings
- Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn)
- Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.)
- Update CLAUDE.md template and hook/workflow templates to use {{packageManager}}
- Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch
- Remove duplicate cost-ops team definition from teams.yaml
- Pin all dependency versions in package.json (remove ^ prefixes)
- Add vitest coverage thresholds (80% statements/branches/functions/lines)
- Fix src/start/ code quality: null guards, exit delay, error boundaries
- Harden consolidate-branches.sh: self-resolution guard, stash restore
- Regenerate all 533 output files via agentkit sync
* chore(engine): add brand color palette variables to sync vars
* fix(ci): fix test race condition, workspace config, and lockfile
- Fix ConversationFlow test: wait for 'Got it' before asserting hint text
- Add packages field to pnpm-workspace.yaml for proper workspace resolution
- Regenerate lockfile after version pinning (removed ^ prefixes)
- Remove accidental .agentkit/templates/src/ artifacts
* fix(start): add event loop yields to ConversationFlow tests
ink-select-input needs setImmediate yields after a new menu renders
before it can process ENTER keypresses. Without this, the second ENTER
in leaf-selection tests was swallowed, causing intermittent failures.
---------
* fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391)
* Add entry point for new framework users (#389)
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* test(start): add coverage for result screen exit path
Add App-level test that navigates through ConversationFlow to a leaf
node and verifies the result screen text is rendered before exit().
Confirms the synchronous useEffect exit is safe — React commits the
render (Ink captures the frame) before useEffect fires.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: update generated sync output timestamps
Generated files updated with current sync date from dev merge.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
---------
* fix: caldues heuristics (#398)
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(quality): resolve all lint and format errors
- Fix Prettier formatting across engine src and start components
- Resolve 381 markdown lint errors (MD040, MD024, MD036, MD029, MD022, MD001)
- Add markdownlint config rules for intentional doc patterns
- Add .claude/worktrees to prettierignore to exclude external branches
- Enable markdownlint MD024/MD026 for duplicate headings and trailing colons
All quality checks now pass: Prettier (0 errors), Markdown lint (0 errors), Tests (111/111), Build (28.9kb)
* I'll update the last_updated field in all the files from '2026-03-12' to '2026-03-13'. This appears to be a routine timestamp update across all the AgentKit Forge generated files.
* docs: add AgentKit Forge sync feedback
Add detailed feedback on AgentKit Forge v3.1.0 integration, focusing on:
- Windows line-ending issues and multi-editor sync behavior
- Documentation scaffold-once limitations and override challenges
- Unresolved placeholder warnings lacking diagnostics
- Windows-specific pnpm execution problems
* docs: update CLAUDE.md with repository-specific editing guidelines
Clarified that modifications to `.agentkit` files are permitted in the agentkit-forge repository, while upstream directories remain protected. This ensures users understand the editing boundaries for project configuration and template files.
* docs(claude): allow .agentkit edits in this repo (NB for framework dev)
Made-with: Cursor
* chore(sync): regenerate outputs after agentkit:sync
Made-with: Cursor
---------
Co-authored-by: Claude <noreply@anthropic.com>
JustAGhosT
added a commit
that referenced
this pull request
Mar 17, 2026
* Add start command: new user entry point with state detection (#387)
* fix(commands): add AskUserQuestion to VALID_TOOLS and /start command
The /start command session got stuck because AskUserQuestion was not
included in the allowed-tools whitelist. This fix addresses three issues:
1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite,
Agent, and NotebookEdit — preventing commands from declaring these
Claude Code built-in tools in their allowed-tools.
2. The /start command template now explicitly includes AskUserQuestion
in its allowed-tools frontmatter and instructs the agent to use it
for interactive guided choices (Phase 3).
3. Added /start command spec to commands.yaml with AskUserQuestion as
a declared tool dependency.
Also adds a test case validating all four newly-added tools are accepted
by the spec validator.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
* fix(commands): remove unrendered Handlebars comment from start.md output
The generated .claude/commands/start.md contained a raw {{! ... }}
Handlebars comment that was not processed by the sync engine. Remove it
so the generated output is clean Markdown.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
---------
Co-authored-by: Claude <noreply@anthropic.com>
* Add configurable prefix to kits commands (#388)
* feat(sync): add configurable command prefix for generated slash commands
Add `commandPrefix` setting to overlay settings that namespaces all
generated slash commands across platforms:
- Claude Code: subdirectory strategy (kits/check.md → /project:kits:check)
- Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check)
- Team commands excluded from prefixing (already namespaced)
Changes:
- Add resolveCommandPath() helper with subdirectory/filename strategies
- Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands,
syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills
- Add commandPrefix to vars from overlay settings
- Add commandPrefixedName to buildCommandVars
- Add 16 unit + integration tests (all pass, 93 existing tests unaffected)
Default is null (no prefix) for full backwards compatibility.
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
* fix(sync): address review findings for command prefix
- Remove unused afterAll import from test file
- Add clarifying comment that non-spec command files are also prefixed
- Add 2 integration tests verifying commandPrefixedName template variable
renders correctly with and without prefix
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(ci): CI remediation — package manager, review findings, test stability (#390)
* fix(ci): resolve 7 bugs from project review
- BUG-001: Replace flaky discover test with controlled temp fixture
- BUG-002: Run prettier --write to fix formatting drift
- BUG-003: Add form-template detection skip in issue label validation
- BUG-005: Change claude.yml to self-hosted runner
- BUG-006: Align branch protection required status checks with project.yaml
- BUG-007: Fix command injection in resolve-merge.sh (use grep -F)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: update changelog, add planning registry review findings
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed)
- Add Project Review Findings section to planning registry (PR-001 to PR-014)
- Update planning docs after sync merge
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(review): add --generate-plans flag to project-review command
Add Phase 2.5 plan generation after project review findings. When
--generate-plans is passed (default: true), scaffold plan files from
critical/high findings into docs/planning/review-findings/.
Also includes sync cleanup of stale cursor/windsurf settings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate all outputs after project review fixes
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cli): dynamic flag loading from commands.yaml + context-aware template hook
- Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags()
that reads flag definitions from commands.yaml at startup
- CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml
- Self-validation warns at startup if any flag is missing a type definition
- Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency
- Fix scaffold-once orphan bug: carry forward manifest entries for files
skipped by scaffold-once so orphan cleanup does not delete them
- Make protect-templates hook context-aware: skip protection in the
agentkit-forge source repo (detected via package.json name) so
maintainer agents can edit templates; block only in downstream repos
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(sync): add managed-mode script templates for downstream repos
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130)
Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating,
validating, and deploying new agent team specifications. Adapted from
cogmesh #130 with a simplified 6-agent pipeline:
- input-clarifier: assess requests, extract constraints
- mission-definer: lock team definition (ID, scope, accepts)
- role-architect: design agent roles and dependencies
- prompt-engineer: write agent descriptions and rules
- flow-designer: design team command and integration points
- team-validator: quality gate for spec consistency
Includes /team-forge command with --task flag (create-team, validate-team,
audit-teams, update-team) and planning doc.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(teams): add Strategic Ops team for cross-project coordination
Add the STRATEGIC OPS team (T12) — handles framework governance,
portfolio analysis, adoption strategy, impact assessment, and release
coordination across all repos using AgentKit Forge.
5-agent pipeline:
- portfolio-analyst: inventory repos, detect drift, adoption metrics
- governance-advisor: versioning strategy, breaking change protocols
- adoption-strategist: onboarding, migration paths, rollout plans
- impact-assessor: blast radius analysis for template/spec changes
- release-coordinator: version bumps, sync waves, release comms
Includes /team-strategic-ops command with --task and --scope flags.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(agents): add agent/team relationship matrix analysis engine + scripts
Add comprehensive agent/team relationship analysis with 8 cross-reference
matrices and 10 supplementary analyses (orphans, cycles, bottlenecks,
reachability, critical path, notification amplifiers, etc.).
- Fix YAML structure: strategic-ops agents now under own top-level key
- Add explicit agents: lists to forge + strategic-ops teams in teams.yaml
- Add consolidation detection responsibilities to portfolio-analyst
- Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers)
- Wire analyze-agents CLI command with --output/--matrix/--format flags
- Add managed-scaffold script templates (bash + PowerShell)
- Integrate into sync pipeline (auto-regenerates matrix on spec changes)
- Add 33 tests covering all matrices, analyses, and edge cases
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: update documentation files and add plan template
- Add trailing newlines to Cursor command documentation files for consistency
- Add new plan template files for project planning
- Improve markdown table formatting in Claude skills documentation
- Remove obsolete .clinerules/testing.md file
- Update various rule files with better formatting and advisory rule alignment
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364)
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365)
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): address CodeRabbit review findings on strategic-ops PR
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent
crash when node_modules is missing
- Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1
templates to fix invalid PowerShell syntax
- Fix protect-templates hook path: use 3 parent traversals to reach
.agentkit/package.json from hooks directory
- Remove schema-invalid 'phase' field from ai-cost-ops rule domain
- Narrow strategic-ops scope: replace **/* catch-all with specific files
- Add strategic-ops to cost-ops handoff-chain for consistency
- Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst,
and grant-hunter agents
- Fix glob matching in resolve-merge.sh (use regex instead of broken
sed strip)
- Add merge failure vs conflict detection in resolve-merge.ps1
- Add branch existence check in setup-agentkit-branch-governance scripts
- Add gh CLI preflight check in sync-split-pr.ps1
- Deduplicate branch protection loop when defaultBranch is 'main'
- Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/**
- Fix non-canonical doc paths in intake-agent-proposal.md
- Add changelog entries for new teams, agents, and analysis engine
- Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope)
- Run prettier on all modified files
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): add gh auth preflight and changelog divider handling
- sync-split-pr.ps1: add gh auth status check before side effects
- update-changelog.ps1: stop before --- divider when appending entries
Addresses CodeRabbit review comments #7 and #13 on PR #356.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(sync): set executable permission on analyze-agents.sh
Linux CI sync produces +x permissions; align local to match.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): address CodeRabbit review round 3-4 findings
- Fix YAML frontmatter in copilot agent template: use double quotes
for description field to handle apostrophes (CRITICAL)
- Fix protect-templates.sh/ps1 path traversal: correct parent
directory count for .agentkit/package.json source repo detection
- Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1
so native command failures (git, pnpm, gh) are treated as fatal
- Replace try/catch with $LASTEXITCODE check for gh auth status
- Deduplicate branch loop in setup-agentkit-branch-governance.sh
when defaultBranch equals 'main'
- Fix duplicate verification echo lines in governance scripts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* style(docs): fix prettier formatting on planning documents
Run prettier --write on web-intake-expansion.md and
intake-agent-proposal.md to fix CI formatting check.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tests): isolate render target gating tests with fresh temp dirs
Tests in the "render target gating" describe block shared a single temp
directory via beforeAll/afterAll. The first test ran --only claude
(generating .claude/ files), and the second ran --only warp expecting no
Claude files — but leftovers from test 1 caused the assertion to fail.
Changing to beforeEach/afterEach gives each test a clean directory.
Closes #377
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(engine): add pre-sync commit guard and interactive apply mode
Adds two safety features to the sync pipeline:
1. Pre-sync commit guard: Detects uncommitted changes in protected
directories (.agentkit/engines, spec, overlays) before sync runs.
In TTY mode, prompts to abort, stash, or continue. In non-TTY
mode (CI), prints a warning and proceeds.
2. Interactive apply mode: After rendering, shows a change summary
and prompts: apply all / skip all / prompt each file. Per-file
prompt supports show-diff and apply-all-remaining. Default in
TTY; bypassed with --yes, --no-prompt, or --force.
New module: sync-guard.mjs with 4 exported functions and 7 tests.
New CLI flags: --yes, --no-prompt for non-interactive sync.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate outputs after branch merges
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate outputs after merge of new-user-entry-point
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(engine): add configurable package manager and fix review findings
- Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn)
- Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.)
- Update CLAUDE.md template and hook/workflow templates to use {{packageManager}}
- Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch
- Remove duplicate cost-ops team definition from teams.yaml
- Pin all dependency versions in package.json (remove ^ prefixes)
- Add vitest coverage thresholds (80% statements/branches/functions/lines)
- Fix src/start/ code quality: null guards, exit delay, error boundaries
- Harden consolidate-branches.sh: self-resolution guard, stash restore
- Regenerate all 533 output files via agentkit sync
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(engine): add brand color palette variables to sync vars
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(ci): fix test race condition, workspace config, and lockfile
- Fix ConversationFlow test: wait for 'Got it' before asserting hint text
- Add packages field to pnpm-workspace.yaml for proper workspace resolution
- Regenerate lockfile after version pinning (removed ^ prefixes)
- Remove accidental .agentkit/templates/src/ artifacts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(start): add event loop yields to ConversationFlow tests
ink-select-input needs setImmediate yields after a new menu renders
before it can process ENTER keypresses. Without this, the second ENTER
in leaf-selection tests was swallowed, causing intermittent failures.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391)
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Add entry point for new framework users (#389)
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* test(start): add coverage for result screen exit path
Add App-level test that navigates through ConversationFlow to a leaf
node and verifies the result screen text is rendered before exit().
Confirms the synchronous useEffect exit is safe — React commits the
render (Ink captures the frame) before useEffect fires.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: update generated sync output timestamps
Generated files updated with current sync date from dev merge.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix: caldues heuristics (#398)
* feat: complete revisit of agents (#399) (#400)
* Add start command: new user entry point with state detection (#387)
* fix(commands): add AskUserQuestion to VALID_TOOLS and /start command
The /start command session got stuck because AskUserQuestion was not
included in the allowed-tools whitelist. This fix addresses three issues:
1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite,
Agent, and NotebookEdit — preventing commands from declaring these
Claude Code built-in tools in their allowed-tools.
2. The /start command template now explicitly includes AskUserQuestion
in its allowed-tools frontmatter and instructs the agent to use it
for interactive guided choices (Phase 3).
3. Added /start command spec to commands.yaml with AskUserQuestion as
a declared tool dependency.
Also adds a test case validating all four newly-added tools are accepted
by the spec validator.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
* fix(commands): remove unrendered Handlebars comment from start.md output
The generated .claude/commands/start.md contained a raw {{! ... }}
Handlebars comment that was not processed by the sync engine. Remove it
so the generated output is clean Markdown.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
---------
* Add configurable prefix to kits commands (#388)
* feat(sync): add configurable command prefix for generated slash commands
Add `commandPrefix` setting to overlay settings that namespaces all
generated slash commands across platforms:
- Claude Code: subdirectory strategy (kits/check.md → /project:kits:check)
- Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check)
- Team commands excluded from prefixing (already namespaced)
Changes:
- Add resolveCommandPath() helper with subdirectory/filename strategies
- Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands,
syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills
- Add commandPrefix to vars from overlay settings
- Add commandPrefixedName to buildCommandVars
- Add 16 unit + integration tests (all pass, 93 existing tests unaffected)
Default is null (no prefix) for full backwards compatibility.
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
* fix(sync): address review findings for command prefix
- Remove unused afterAll import from test file
- Add clarifying comment that non-spec command files are also prefixed
- Add 2 integration tests verifying commandPrefixedName template variable
renders correctly with and without prefix
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
---------
* fix(ci): CI remediation — package manager, review findings, test stability (#390)
* fix(ci): resolve 7 bugs from project review
- BUG-001: Replace flaky discover test with controlled temp fixture
- BUG-002: Run prettier --write to fix formatting drift
- BUG-003: Add form-template detection skip in issue label validation
- BUG-005: Change claude.yml to self-hosted runner
- BUG-006: Align branch protection required status checks with project.yaml
- BUG-007: Fix command injection in resolve-merge.sh (use grep -F)
* docs: update changelog, add planning registry review findings
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed)
- Add Project Review Findings section to planning registry (PR-001 to PR-014)
- Update planning docs after sync merge
* feat(review): add --generate-plans flag to project-review command
Add Phase 2.5 plan generation after project review findings. When
--generate-plans is passed (default: true), scaffold plan files from
critical/high findings into docs/planning/review-findings/.
Also includes sync cleanup of stale cursor/windsurf settings.
* chore(sync): regenerate all outputs after project review fixes
* feat(cli): dynamic flag loading from commands.yaml + context-aware template hook
- Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags()
that reads flag definitions from commands.yaml at startup
- CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml
- Self-validation warns at startup if any flag is missing a type definition
- Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency
- Fix scaffold-once orphan bug: carry forward manifest entries for files
skipped by scaffold-once so orphan cleanup does not delete them
- Make protect-templates hook context-aware: skip protection in the
agentkit-forge source repo (detected via package.json name) so
maintainer agents can edit templates; block only in downstream repos
* feat(sync): add managed-mode script templates for downstream repos
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
* feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130)
Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating,
validating, and deploying new agent team specifications. Adapted from
cogmesh #130 with a simplified 6-agent pipeline:
- input-clarifier: assess requests, extract constraints
- mission-definer: lock team definition (ID, scope, accepts)
- role-architect: design agent roles and dependencies
- prompt-engineer: write agent descriptions and rules
- flow-designer: design team command and integration points
- team-validator: quality gate for spec consistency
Includes /team-forge command with --task flag (create-team, validate-team,
audit-teams, update-team) and planning doc.
* feat(teams): add Strategic Ops team for cross-project coordination
Add the STRATEGIC OPS team (T12) — handles framework governance,
portfolio analysis, adoption strategy, impact assessment, and release
coordination across all repos using AgentKit Forge.
5-agent pipeline:
- portfolio-analyst: inventory repos, detect drift, adoption metrics
- governance-advisor: versioning strategy, breaking change protocols
- adoption-strategist: onboarding, migration paths, rollout plans
- impact-assessor: blast radius analysis for template/spec changes
- release-coordinator: version bumps, sync waves, release comms
Includes /team-strategic-ops command with --task and --scope flags.
* feat(agents): add agent/team relationship matrix analysis engine + scripts
Add comprehensive agent/team relationship analysis with 8 cross-reference
matrices and 10 supplementary analyses (orphans, cycles, bottlenecks,
reachability, critical path, notification amplifiers, etc.).
- Fix YAML structure: strategic-ops agents now under own top-level key
- Add explicit agents: lists to forge + strategic-ops teams in teams.yaml
- Add consolidation detection responsibilities to portfolio-analyst
- Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers)
- Wire analyze-agents CLI command with --output/--matrix/--format flags
- Add managed-scaffold script templates (bash + PowerShell)
- Integrate into sync pipeline (auto-regenerates matrix on spec changes)
- Add 33 tests covering all matrices, analyses, and edge cases
* chore: update documentation files and add plan template
- Add trailing newlines to Cursor command documentation files for consistency
- Add new plan template files for project planning
- Improve markdown table formatting in Claude skills documentation
- Remove obsolete .clinerules/testing.md file
- Update various rule files with better formatting and advisory rule alignment
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
* feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364)
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365)
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
---------
* fix(teams): address CodeRabbit review findings on strategic-ops PR
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent
crash when node_modules is missing
- Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1
templates to fix invalid PowerShell syntax
- Fix protect-templates hook path: use 3 parent traversals to reach
.agentkit/package.json from hooks directory
- Remove schema-invalid 'phase' field from ai-cost-ops rule domain
- Narrow strategic-ops scope: replace **/* catch-all with specific files
- Add strategic-ops to cost-ops handoff-chain for consistency
- Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst,
and grant-hunter agents
- Fix glob matching in resolve-merge.sh (use regex instead of broken
sed strip)
- Add merge failure vs conflict detection in resolve-merge.ps1
- Add branch existence check in setup-agentkit-branch-governance scripts
- Add gh CLI preflight check in sync-split-pr.ps1
- Deduplicate branch protection loop when defaultBranch is 'main'
- Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/**
- Fix non-canonical doc paths in intake-agent-proposal.md
- Add changelog entries for new teams, agents, and analysis engine
- Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope)
- Run prettier on all modified files
* fix(templates): add gh auth preflight and changelog divider handling
- sync-split-pr.ps1: add gh auth status check before side effects
- update-changelog.ps1: stop before --- divider when appending entries
Addresses CodeRabbit review comments #7 and #13 on PR #356.
* fix(sync): set executable permission on analyze-agents.sh
Linux CI sync produces +x permissions; align local to match.
* fix(templates): address CodeRabbit review round 3-4 findings
- Fix YAML frontmatter in copilot agent template: use double quotes
for description field to handle apostrophes (CRITICAL)
- Fix protect-templates.sh/ps1 path traversal: correct parent
directory count for .agentkit/package.json source repo detection
- Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1
so native command failures (git, pnpm, gh) are treated as fatal
- Replace try/catch with $LASTEXITCODE check for gh auth status
- Deduplicate branch loop in setup-agentkit-branch-governance.sh
when defaultBranch equals 'main'
- Fix duplicate verification echo lines in governance scripts
* style(docs): fix prettier formatting on planning documents
Run prettier --write on web-intake-expansion.md and
intake-agent-proposal.md to fix CI formatting check.
* fix(tests): isolate render target gating tests with fresh temp dirs
Tests in the "render target gating" describe block shared a single temp
directory via beforeAll/afterAll. The first test ran --only claude
(generating .claude/ files), and the second ran --only warp expecting no
Claude files — but leftovers from test 1 caused the assertion to fail.
Changing to beforeEach/afterEach gives each test a clean directory.
Closes #377
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(engine): add pre-sync commit guard and interactive apply mode
Adds two safety features to the sync pipeline:
1. Pre-sync commit guard: Detects uncommitted changes in protected
directories (.agentkit/engines, spec, overlays) before sync runs.
In TTY mode, prompts to abort, stash, or continue. In non-TTY
mode (CI), prints a warning and proceeds.
2. Interactive apply mode: After rendering, shows a change summary
and prompts: apply all / skip all / prompt each file. Per-file
prompt supports show-diff and apply-all-remaining. Default in
TTY; bypassed with --yes, --no-prompt, or --force.
New module: sync-guard.mjs with 4 exported functions and 7 tests.
New CLI flags: --yes, --no-prompt for non-interactive sync.
* chore(sync): regenerate outputs after branch merges
* chore(sync): regenerate outputs after merge of new-user-entry-point
* feat(engine): add configurable package manager and fix review findings
- Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn)
- Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.)
- Update CLAUDE.md template and hook/workflow templates to use {{packageManager}}
- Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch
- Remove duplicate cost-ops team definition from teams.yaml
- Pin all dependency versions in package.json (remove ^ prefixes)
- Add vitest coverage thresholds (80% statements/branches/functions/lines)
- Fix src/start/ code quality: null guards, exit delay, error boundaries
- Harden consolidate-branches.sh: self-resolution guard, stash restore
- Regenerate all 533 output files via agentkit sync
* chore(engine): add brand color palette variables to sync vars
* fix(ci): fix test race condition, workspace config, and lockfile
- Fix ConversationFlow test: wait for 'Got it' before asserting hint text
- Add packages field to pnpm-workspace.yaml for proper workspace resolution
- Regenerate lockfile after version pinning (removed ^ prefixes)
- Remove accidental .agentkit/templates/src/ artifacts
* fix(start): add event loop yields to ConversationFlow tests
ink-select-input needs setImmediate yields after a new menu renders
before it can process ENTER keypresses. Without this, the second ENTER
in leaf-selection tests was swallowed, causing intermittent failures.
---------
* fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391)
* Add entry point for new framework users (#389)
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* test(start): add coverage for result screen exit path
Add App-level test that navigates through ConversationFlow to a leaf
node and verifies the result screen text is rendered before exit().
Confirms the synchronous useEffect exit is safe — React commits the
render (Ink captures the frame) before useEffect fires.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: update generated sync output timestamps
Generated files updated with current sync date from dev merge.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
---------
* fix: caldues heuristics (#398)
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(quality): resolve all lint and format errors
- Fix Prettier formatting across engine src and start components
- Resolve 381 markdown lint errors (MD040, MD024, MD036, MD029, MD022, MD001)
- Add markdownlint config rules for intentional doc patterns
- Add .claude/worktrees to prettierignore to exclude external branches
- Enable markdownlint MD024/MD026 for duplicate headings and trailing colons
All quality checks now pass: Prettier (0 errors), Markdown lint (0 errors), Tests (111/111), Build (28.9kb)
* I'll update the last_updated field in all the files from '2026-03-12' to '2026-03-13'. This appears to be a routine timestamp update across all the AgentKit Forge generated files.
* docs: add AgentKit Forge sync feedback
Add detailed feedback on AgentKit Forge v3.1.0 integration, focusing on:
- Windows line-ending issues and multi-editor sync behavior
- Documentation scaffold-once limitations and override challenges
- Unresolved placeholder warnings lacking diagnostics
- Windows-specific pnpm execution problems
* docs: update CLAUDE.md with repository-specific editing guidelines
Clarified that modifications to `.agentkit` files are permitted in the agentkit-forge repository, while upstream directories remain protected. This ensures users understand the editing boundaries for project configuration and template files.
* docs(claude): allow .agentkit edits in this repo (NB for framework dev)
Made-with: Cursor
* chore(sync): regenerate outputs after agentkit:sync
Made-with: Cursor
* chore(sync): regenerate outputs and update timestamps
Sync engine regenerated all platform outputs with updated timestamps.
Removed yarn.lock references from resolve-merge.sh.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Fix/generated files and conflict markers (#427)
* fix(infra): resolve container app fqdn attribute and format code
* chore(sync): update AGENT_BACKLOG.md and other files for task management
- Enhanced AGENT_BACKLOG.md with detailed task scopes for CI pipeline configuration and test framework setup.
- Added new docker-compose.yml for local/staging validation of the framework.
- Updated CONTRIBUTING.md to include documentation hub link in the Discovery phase.
- Introduced README.md files in db, infra, and migrations directories to clarify their purpose for adopters.
- Added API conventions documentation to guide adopters on structuring their APIs.
- Created implementation plan for state management improvements and added relevant tests.
- Regenerated outputs across various files to reflect recent changes and ensure consistency.
* chore(sync): regenerate outputs after dev merge
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude <noreply@anthropic.com>
JustAGhosT
added a commit
that referenced
this pull request
Mar 17, 2026
…cord (#431) * feat: new entry points (#426) * Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- Co-authored-by: Claude <noreply@anthropic.com> * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills - Add commandPrefix to vars from overlay settings - Add commandPrefixedName to buildCommandVars - Add 16 unit + integration tests (all pass, 93 existing tests unaffected) Default is null (no prefix) for full backwards compatibility. https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p * fix(sync): address review findings for command prefix - Remove unused afterAll import from test file - Add clarifying comment that non-spec command files are also prefixed - Add 2 integration tests verifying commandPrefixedName template variable renders correctly with and without prefix https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p --------- Co-authored-by: Claude <noreply@anthropic.com> * fix(ci): CI remediation — package manager, review findings, test stability (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate all outputs after project review fixes Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate outputs after branch merges Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate outputs after merge of new-user-entry-point Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(engine): add brand color palette variables to sync vars Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391) Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * Add entry point for new framework users (#389) * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * test(start): add coverage for result screen exit path Add App-level test that navigates through ConversationFlow to a leaf node and verifies the result screen text is rendered before exit(). Confirms the synchronous useEffect exit is safe — React commits the render (Ink captures the frame) before useEffect fires. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: update generated sync output timestamps Generated files updated with current sync date from dev merge. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW --------- Co-authored-by: Claude <noreply@anthropic.com> * fix: caldues heuristics (#398) * feat: complete revisit of agents (#399) (#400) * Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills - Add commandPrefix to vars from overlay settings - Add commandPrefixedName to buildCommandVars - Add 16 unit + integration tests (all pass, 93 existing tests unaffected) Default is null (no prefix) for full backwards compatibility. https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p * fix(sync): address review findings for command prefix - Remove unused afterAll import from test file - Add clarifying comment that non-spec command files are also prefixed - Add 2 integration tests verifying commandPrefixedName template variable renders correctly with and without prefix https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p --------- * fix(ci): CI remediation — package manager, review findings, test stability (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. * chore(sync): regenerate all outputs after project review fixes * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs --------- * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. * chore(sync): regenerate outputs after branch merges * chore(sync): regenerate outputs after merge of new-user-entry-point * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync * chore(engine): add brand color palette variables to sync vars * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. --------- * fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391) * Add entry point for new framework users (#389) * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * test(start): add coverage for result screen exit path Add App-level test that navigates through ConversationFlow to a leaf node and verifies the result screen text is rendered before exit(). Confirms the synchronous useEffect exit is safe — React commits the render (Ink captures the frame) before useEffect fires. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: update generated sync output timestamps Generated files updated with current sync date from dev merge. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW --------- * fix: caldues heuristics (#398) --------- Co-authored-by: Claude <noreply@anthropic.com> * fix(quality): resolve all lint and format errors - Fix Prettier formatting across engine src and start components - Resolve 381 markdown lint errors (MD040, MD024, MD036, MD029, MD022, MD001) - Add markdownlint config rules for intentional doc patterns - Add .claude/worktrees to prettierignore to exclude external branches - Enable markdownlint MD024/MD026 for duplicate headings and trailing colons All quality checks now pass: Prettier (0 errors), Markdown lint (0 errors), Tests (111/111), Build (28.9kb) * I'll update the last_updated field in all the files from '2026-03-12' to '2026-03-13'. This appears to be a routine timestamp update across all the AgentKit Forge generated files. * docs: add AgentKit Forge sync feedback Add detailed feedback on AgentKit Forge v3.1.0 integration, focusing on: - Windows line-ending issues and multi-editor sync behavior - Documentation scaffold-once limitations and override challenges - Unresolved placeholder warnings lacking diagnostics - Windows-specific pnpm execution problems * docs: update CLAUDE.md with repository-specific editing guidelines Clarified that modifications to `.agentkit` files are permitted in the agentkit-forge repository, while upstream directories remain protected. This ensures users understand the editing boundaries for project configuration and template files. * docs(claude): allow .agentkit edits in this repo (NB for framework dev) Made-with: Cursor * chore(sync): regenerate outputs after agentkit:sync Made-with: Cursor --------- Co-authored-by: Claude <noreply@anthropic.com> * chore(sync): regenerate outputs and update timestamps (#429) * Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- Co-authored-by: Claude <noreply@anthropic.com> * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts…
JustAGhosT
added a commit
that referenced
this pull request
Mar 21, 2026
…#432) * feat: new entry points (#426) * Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- Co-authored-by: Claude <noreply@anthropic.com> * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills - Add commandPrefix to vars from overlay settings - Add commandPrefixedName to buildCommandVars - Add 16 unit + integration tests (all pass, 93 existing tests unaffected) Default is null (no prefix) for full backwards compatibility. https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p * fix(sync): address review findings for command prefix - Remove unused afterAll import from test file - Add clarifying comment that non-spec command files are also prefixed - Add 2 integration tests verifying commandPrefixedName template variable renders correctly with and without prefix https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p --------- Co-authored-by: Claude <noreply@anthropic.com> * fix(ci): CI remediation — package manager, review findings, test stability (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate all outputs after project review fixes Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate outputs after branch merges Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate outputs after merge of new-user-entry-point Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(engine): add brand color palette variables to sync vars Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391) Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * Add entry point for new framework users (#389) * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * test(start): add coverage for result screen exit path Add App-level test that navigates through ConversationFlow to a leaf node and verifies the result screen text is rendered before exit(). Confirms the synchronous useEffect exit is safe — React commits the render (Ink captures the frame) before useEffect fires. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: update generated sync output timestamps Generated files updated with current sync date from dev merge. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW --------- Co-authored-by: Claude <noreply@anthropic.com> * fix: caldues heuristics (#398) * feat: complete revisit of agents (#399) (#400) * Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills - Add commandPrefix to vars from overlay settings - Add commandPrefixedName to buildCommandVars - Add 16 unit + integration tests (all pass, 93 existing tests unaffected) Default is null (no prefix) for full backwards compatibility. https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p * fix(sync): address review findings for command prefix - Remove unused afterAll import from test file - Add clarifying comment that non-spec command files are also prefixed - Add 2 integration tests verifying commandPrefixedName template variable renders correctly with and without prefix https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p --------- * fix(ci): CI remediation — package manager, review findings, test stability (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. * chore(sync): regenerate all outputs after project review fixes * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs --------- * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. * chore(sync): regenerate outputs after branch merges * chore(sync): regenerate outputs after merge of new-user-entry-point * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync * chore(engine): add brand color palette variables to sync vars * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. --------- * fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391) * Add entry point for new framework users (#389) * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * test(start): add coverage for result screen exit path Add App-level test that navigates through ConversationFlow to a leaf node and verifies the result screen text is rendered before exit(). Confirms the synchronous useEffect exit is safe — React commits the render (Ink captures the frame) before useEffect fires. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: update generated sync output timestamps Generated files updated with current sync date from dev merge. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW --------- * fix: caldues heuristics (#398) --------- Co-authored-by: Claude <noreply@anthropic.com> * fix(quality): resolve all lint and format errors - Fix Prettier formatting across engine src and start components - Resolve 381 markdown lint errors (MD040, MD024, MD036, MD029, MD022, MD001) - Add markdownlint config rules for intentional doc patterns - Add .claude/worktrees to prettierignore to exclude external branches - Enable markdownlint MD024/MD026 for duplicate headings and trailing colons All quality checks now pass: Prettier (0 errors), Markdown lint (0 errors), Tests (111/111), Build (28.9kb) * I'll update the last_updated field in all the files from '2026-03-12' to '2026-03-13'. This appears to be a routine timestamp update across all the AgentKit Forge generated files. * docs: add AgentKit Forge sync feedback Add detailed feedback on AgentKit Forge v3.1.0 integration, focusing on: - Windows line-ending issues and multi-editor sync behavior - Documentation scaffold-once limitations and override challenges - Unresolved placeholder warnings lacking diagnostics - Windows-specific pnpm execution problems * docs: update CLAUDE.md with repository-specific editing guidelines Clarified that modifications to `.agentkit` files are permitted in the agentkit-forge repository, while upstream directories remain protected. This ensures users understand the editing boundaries for project configuration and template files. * docs(claude): allow .agentkit edits in this repo (NB for framework dev) Made-with: Cursor * chore(sync): regenerate outputs after agentkit:sync Made-with: Cursor --------- Co-authored-by: Claude <noreply@anthropic.com> * chore(sync): regenerate outputs and update timestamps (#429) * Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- Co-authored-by: Claude <noreply@anthropic.com> * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, sy…
JustAGhosT
added a commit
that referenced
this pull request
Mar 21, 2026
* Add start command: new user entry point with state detection (#387)
* fix(commands): add AskUserQuestion to VALID_TOOLS and /start command
The /start command session got stuck because AskUserQuestion was not
included in the allowed-tools whitelist. This fix addresses three issues:
1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite,
Agent, and NotebookEdit — preventing commands from declaring these
Claude Code built-in tools in their allowed-tools.
2. The /start command template now explicitly includes AskUserQuestion
in its allowed-tools frontmatter and instructs the agent to use it
for interactive guided choices (Phase 3).
3. Added /start command spec to commands.yaml with AskUserQuestion as
a declared tool dependency.
Also adds a test case validating all four newly-added tools are accepted
by the spec validator.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
* fix(commands): remove unrendered Handlebars comment from start.md output
The generated .claude/commands/start.md contained a raw {{! ... }}
Handlebars comment that was not processed by the sync engine. Remove it
so the generated output is clean Markdown.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
---------
Co-authored-by: Claude <noreply@anthropic.com>
* Add configurable prefix to kits commands (#388)
* feat(sync): add configurable command prefix for generated slash commands
Add `commandPrefix` setting to overlay settings that namespaces all
generated slash commands across platforms:
- Claude Code: subdirectory strategy (kits/check.md → /project:kits:check)
- Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check)
- Team commands excluded from prefixing (already namespaced)
Changes:
- Add resolveCommandPath() helper with subdirectory/filename strategies
- Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands,
syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills
- Add commandPrefix to vars from overlay settings
- Add commandPrefixedName to buildCommandVars
- Add 16 unit + integration tests (all pass, 93 existing tests unaffected)
Default is null (no prefix) for full backwards compatibility.
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
* fix(sync): address review findings for command prefix
- Remove unused afterAll import from test file
- Add clarifying comment that non-spec command files are also prefixed
- Add 2 integration tests verifying commandPrefixedName template variable
renders correctly with and without prefix
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(ci): CI remediation — package manager, review findings, test stability (#390)
* fix(ci): resolve 7 bugs from project review
- BUG-001: Replace flaky discover test with controlled temp fixture
- BUG-002: Run prettier --write to fix formatting drift
- BUG-003: Add form-template detection skip in issue label validation
- BUG-005: Change claude.yml to self-hosted runner
- BUG-006: Align branch protection required status checks with project.yaml
- BUG-007: Fix command injection in resolve-merge.sh (use grep -F)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: update changelog, add planning registry review findings
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed)
- Add Project Review Findings section to planning registry (PR-001 to PR-014)
- Update planning docs after sync merge
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(review): add --generate-plans flag to project-review command
Add Phase 2.5 plan generation after project review findings. When
--generate-plans is passed (default: true), scaffold plan files from
critical/high findings into docs/planning/review-findings/.
Also includes sync cleanup of stale cursor/windsurf settings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate all outputs after project review fixes
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cli): dynamic flag loading from commands.yaml + context-aware template hook
- Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags()
that reads flag definitions from commands.yaml at startup
- CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml
- Self-validation warns at startup if any flag is missing a type definition
- Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency
- Fix scaffold-once orphan bug: carry forward manifest entries for files
skipped by scaffold-once so orphan cleanup does not delete them
- Make protect-templates hook context-aware: skip protection in the
agentkit-forge source repo (detected via package.json name) so
maintainer agents can edit templates; block only in downstream repos
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(sync): add managed-mode script templates for downstream repos
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130)
Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating,
validating, and deploying new agent team specifications. Adapted from
cogmesh #130 with a simplified 6-agent pipeline:
- input-clarifier: assess requests, extract constraints
- mission-definer: lock team definition (ID, scope, accepts)
- role-architect: design agent roles and dependencies
- prompt-engineer: write agent descriptions and rules
- flow-designer: design team command and integration points
- team-validator: quality gate for spec consistency
Includes /team-forge command with --task flag (create-team, validate-team,
audit-teams, update-team) and planning doc.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(teams): add Strategic Ops team for cross-project coordination
Add the STRATEGIC OPS team (T12) — handles framework governance,
portfolio analysis, adoption strategy, impact assessment, and release
coordination across all repos using AgentKit Forge.
5-agent pipeline:
- portfolio-analyst: inventory repos, detect drift, adoption metrics
- governance-advisor: versioning strategy, breaking change protocols
- adoption-strategist: onboarding, migration paths, rollout plans
- impact-assessor: blast radius analysis for template/spec changes
- release-coordinator: version bumps, sync waves, release comms
Includes /team-strategic-ops command with --task and --scope flags.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(agents): add agent/team relationship matrix analysis engine + scripts
Add comprehensive agent/team relationship analysis with 8 cross-reference
matrices and 10 supplementary analyses (orphans, cycles, bottlenecks,
reachability, critical path, notification amplifiers, etc.).
- Fix YAML structure: strategic-ops agents now under own top-level key
- Add explicit agents: lists to forge + strategic-ops teams in teams.yaml
- Add consolidation detection responsibilities to portfolio-analyst
- Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers)
- Wire analyze-agents CLI command with --output/--matrix/--format flags
- Add managed-scaffold script templates (bash + PowerShell)
- Integrate into sync pipeline (auto-regenerates matrix on spec changes)
- Add 33 tests covering all matrices, analyses, and edge cases
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: update documentation files and add plan template
- Add trailing newlines to Cursor command documentation files for consistency
- Add new plan template files for project planning
- Improve markdown table formatting in Claude skills documentation
- Remove obsolete .clinerules/testing.md file
- Update various rule files with better formatting and advisory rule alignment
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364)
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365)
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): address CodeRabbit review findings on strategic-ops PR
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent
crash when node_modules is missing
- Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1
templates to fix invalid PowerShell syntax
- Fix protect-templates hook path: use 3 parent traversals to reach
.agentkit/package.json from hooks directory
- Remove schema-invalid 'phase' field from ai-cost-ops rule domain
- Narrow strategic-ops scope: replace **/* catch-all with specific files
- Add strategic-ops to cost-ops handoff-chain for consistency
- Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst,
and grant-hunter agents
- Fix glob matching in resolve-merge.sh (use regex instead of broken
sed strip)
- Add merge failure vs conflict detection in resolve-merge.ps1
- Add branch existence check in setup-agentkit-branch-governance scripts
- Add gh CLI preflight check in sync-split-pr.ps1
- Deduplicate branch protection loop when defaultBranch is 'main'
- Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/**
- Fix non-canonical doc paths in intake-agent-proposal.md
- Add changelog entries for new teams, agents, and analysis engine
- Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope)
- Run prettier on all modified files
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): add gh auth preflight and changelog divider handling
- sync-split-pr.ps1: add gh auth status check before side effects
- update-changelog.ps1: stop before --- divider when appending entries
Addresses CodeRabbit review comments #7 and #13 on PR #356.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(sync): set executable permission on analyze-agents.sh
Linux CI sync produces +x permissions; align local to match.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): address CodeRabbit review round 3-4 findings
- Fix YAML frontmatter in copilot agent template: use double quotes
for description field to handle apostrophes (CRITICAL)
- Fix protect-templates.sh/ps1 path traversal: correct parent
directory count for .agentkit/package.json source repo detection
- Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1
so native command failures (git, pnpm, gh) are treated as fatal
- Replace try/catch with $LASTEXITCODE check for gh auth status
- Deduplicate branch loop in setup-agentkit-branch-governance.sh
when defaultBranch equals 'main'
- Fix duplicate verification echo lines in governance scripts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* style(docs): fix prettier formatting on planning documents
Run prettier --write on web-intake-expansion.md and
intake-agent-proposal.md to fix CI formatting check.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tests): isolate render target gating tests with fresh temp dirs
Tests in the "render target gating" describe block shared a single temp
directory via beforeAll/afterAll. The first test ran --only claude
(generating .claude/ files), and the second ran --only warp expecting no
Claude files — but leftovers from test 1 caused the assertion to fail.
Changing to beforeEach/afterEach gives each test a clean directory.
Closes #377
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(engine): add pre-sync commit guard and interactive apply mode
Adds two safety features to the sync pipeline:
1. Pre-sync commit guard: Detects uncommitted changes in protected
directories (.agentkit/engines, spec, overlays) before sync runs.
In TTY mode, prompts to abort, stash, or continue. In non-TTY
mode (CI), prints a warning and proceeds.
2. Interactive apply mode: After rendering, shows a change summary
and prompts: apply all / skip all / prompt each file. Per-file
prompt supports show-diff and apply-all-remaining. Default in
TTY; bypassed with --yes, --no-prompt, or --force.
New module: sync-guard.mjs with 4 exported functions and 7 tests.
New CLI flags: --yes, --no-prompt for non-interactive sync.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate outputs after branch merges
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate outputs after merge of new-user-entry-point
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(engine): add configurable package manager and fix review findings
- Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn)
- Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.)
- Update CLAUDE.md template and hook/workflow templates to use {{packageManager}}
- Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch
- Remove duplicate cost-ops team definition from teams.yaml
- Pin all dependency versions in package.json (remove ^ prefixes)
- Add vitest coverage thresholds (80% statements/branches/functions/lines)
- Fix src/start/ code quality: null guards, exit delay, error boundaries
- Harden consolidate-branches.sh: self-resolution guard, stash restore
- Regenerate all 533 output files via agentkit sync
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(engine): add brand color palette variables to sync vars
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(ci): fix test race condition, workspace config, and lockfile
- Fix ConversationFlow test: wait for 'Got it' before asserting hint text
- Add packages field to pnpm-workspace.yaml for proper workspace resolution
- Regenerate lockfile after version pinning (removed ^ prefixes)
- Remove accidental .agentkit/templates/src/ artifacts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(start): add event loop yields to ConversationFlow tests
ink-select-input needs setImmediate yields after a new menu renders
before it can process ENTER keypresses. Without this, the second ENTER
in leaf-selection tests was swallowed, causing intermittent failures.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391)
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Add entry point for new framework users (#389)
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* test(start): add coverage for result screen exit path
Add App-level test that navigates through ConversationFlow to a leaf
node and verifies the result screen text is rendered before exit().
Confirms the synchronous useEffect exit is safe — React commits the
render (Ink captures the frame) before useEffect fires.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: update generated sync output timestamps
Generated files updated with current sync date from dev merge.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix: caldues heuristics (#398)
* feat: complete revisit of agents (#399) (#400)
* Add start command: new user entry point with state detection (#387)
* fix(commands): add AskUserQuestion to VALID_TOOLS and /start command
The /start command session got stuck because AskUserQuestion was not
included in the allowed-tools whitelist. This fix addresses three issues:
1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite,
Agent, and NotebookEdit — preventing commands from declaring these
Claude Code built-in tools in their allowed-tools.
2. The /start command template now explicitly includes AskUserQuestion
in its allowed-tools frontmatter and instructs the agent to use it
for interactive guided choices (Phase 3).
3. Added /start command spec to commands.yaml with AskUserQuestion as
a declared tool dependency.
Also adds a test case validating all four newly-added tools are accepted
by the spec validator.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
* fix(commands): remove unrendered Handlebars comment from start.md output
The generated .claude/commands/start.md contained a raw {{! ... }}
Handlebars comment that was not processed by the sync engine. Remove it
so the generated output is clean Markdown.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
---------
* Add configurable prefix to kits commands (#388)
* feat(sync): add configurable command prefix for generated slash commands
Add `commandPrefix` setting to overlay settings that namespaces all
generated slash commands across platforms:
- Claude Code: subdirectory strategy (kits/check.md → /project:kits:check)
- Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check)
- Team commands excluded from prefixing (already namespaced)
Changes:
- Add resolveCommandPath() helper with subdirectory/filename strategies
- Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands,
syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills
- Add commandPrefix to vars from overlay settings
- Add commandPrefixedName to buildCommandVars
- Add 16 unit + integration tests (all pass, 93 existing tests unaffected)
Default is null (no prefix) for full backwards compatibility.
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
* fix(sync): address review findings for command prefix
- Remove unused afterAll import from test file
- Add clarifying comment that non-spec command files are also prefixed
- Add 2 integration tests verifying commandPrefixedName template variable
renders correctly with and without prefix
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
---------
* fix(ci): CI remediation — package manager, review findings, test stability (#390)
* fix(ci): resolve 7 bugs from project review
- BUG-001: Replace flaky discover test with controlled temp fixture
- BUG-002: Run prettier --write to fix formatting drift
- BUG-003: Add form-template detection skip in issue label validation
- BUG-005: Change claude.yml to self-hosted runner
- BUG-006: Align branch protection required status checks with project.yaml
- BUG-007: Fix command injection in resolve-merge.sh (use grep -F)
* docs: update changelog, add planning registry review findings
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed)
- Add Project Review Findings section to planning registry (PR-001 to PR-014)
- Update planning docs after sync merge
* feat(review): add --generate-plans flag to project-review command
Add Phase 2.5 plan generation after project review findings. When
--generate-plans is passed (default: true), scaffold plan files from
critical/high findings into docs/planning/review-findings/.
Also includes sync cleanup of stale cursor/windsurf settings.
* chore(sync): regenerate all outputs after project review fixes
* feat(cli): dynamic flag loading from commands.yaml + context-aware template hook
- Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags()
that reads flag definitions from commands.yaml at startup
- CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml
- Self-validation warns at startup if any flag is missing a type definition
- Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency
- Fix scaffold-once orphan bug: carry forward manifest entries for files
skipped by scaffold-once so orphan cleanup does not delete them
- Make protect-templates hook context-aware: skip protection in the
agentkit-forge source repo (detected via package.json name) so
maintainer agents can edit templates; block only in downstream repos
* feat(sync): add managed-mode script templates for downstream repos
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
* feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130)
Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating,
validating, and deploying new agent team specifications. Adapted from
cogmesh #130 with a simplified 6-agent pipeline:
- input-clarifier: assess requests, extract constraints
- mission-definer: lock team definition (ID, scope, accepts)
- role-architect: design agent roles and dependencies
- prompt-engineer: write agent descriptions and rules
- flow-designer: design team command and integration points
- team-validator: quality gate for spec consistency
Includes /team-forge command with --task flag (create-team, validate-team,
audit-teams, update-team) and planning doc.
* feat(teams): add Strategic Ops team for cross-project coordination
Add the STRATEGIC OPS team (T12) — handles framework governance,
portfolio analysis, adoption strategy, impact assessment, and release
coordination across all repos using AgentKit Forge.
5-agent pipeline:
- portfolio-analyst: inventory repos, detect drift, adoption metrics
- governance-advisor: versioning strategy, breaking change protocols
- adoption-strategist: onboarding, migration paths, rollout plans
- impact-assessor: blast radius analysis for template/spec changes
- release-coordinator: version bumps, sync waves, release comms
Includes /team-strategic-ops command with --task and --scope flags.
* feat(agents): add agent/team relationship matrix analysis engine + scripts
Add comprehensive agent/team relationship analysis with 8 cross-reference
matrices and 10 supplementary analyses (orphans, cycles, bottlenecks,
reachability, critical path, notification amplifiers, etc.).
- Fix YAML structure: strategic-ops agents now under own top-level key
- Add explicit agents: lists to forge + strategic-ops teams in teams.yaml
- Add consolidation detection responsibilities to portfolio-analyst
- Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers)
- Wire analyze-agents CLI command with --output/--matrix/--format flags
- Add managed-scaffold script templates (bash + PowerShell)
- Integrate into sync pipeline (auto-regenerates matrix on spec changes)
- Add 33 tests covering all matrices, analyses, and edge cases
* chore: update documentation files and add plan template
- Add trailing newlines to Cursor command documentation files for consistency
- Add new plan template files for project planning
- Improve markdown table formatting in Claude skills documentation
- Remove obsolete .clinerules/testing.md file
- Update various rule files with better formatting and advisory rule alignment
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
* feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364)
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365)
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
---------
* fix(teams): address CodeRabbit review findings on strategic-ops PR
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent
crash when node_modules is missing
- Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1
templates to fix invalid PowerShell syntax
- Fix protect-templates hook path: use 3 parent traversals to reach
.agentkit/package.json from hooks directory
- Remove schema-invalid 'phase' field from ai-cost-ops rule domain
- Narrow strategic-ops scope: replace **/* catch-all with specific files
- Add strategic-ops to cost-ops handoff-chain for consistency
- Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst,
and grant-hunter agents
- Fix glob matching in resolve-merge.sh (use regex instead of broken
sed strip)
- Add merge failure vs conflict detection in resolve-merge.ps1
- Add branch existence check in setup-agentkit-branch-governance scripts
- Add gh CLI preflight check in sync-split-pr.ps1
- Deduplicate branch protection loop when defaultBranch is 'main'
- Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/**
- Fix non-canonical doc paths in intake-agent-proposal.md
- Add changelog entries for new teams, agents, and analysis engine
- Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope)
- Run prettier on all modified files
* fix(templates): add gh auth preflight and changelog divider handling
- sync-split-pr.ps1: add gh auth status check before side effects
- update-changelog.ps1: stop before --- divider when appending entries
Addresses CodeRabbit review comments #7 and #13 on PR #356.
* fix(sync): set executable permission on analyze-agents.sh
Linux CI sync produces +x permissions; align local to match.
* fix(templates): address CodeRabbit review round 3-4 findings
- Fix YAML frontmatter in copilot agent template: use double quotes
for description field to handle apostrophes (CRITICAL)
- Fix protect-templates.sh/ps1 path traversal: correct parent
directory count for .agentkit/package.json source repo detection
- Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1
so native command failures (git, pnpm, gh) are treated as fatal
- Replace try/catch with $LASTEXITCODE check for gh auth status
- Deduplicate branch loop in setup-agentkit-branch-governance.sh
when defaultBranch equals 'main'
- Fix duplicate verification echo lines in governance scripts
* style(docs): fix prettier formatting on planning documents
Run prettier --write on web-intake-expansion.md and
intake-agent-proposal.md to fix CI formatting check.
* fix(tests): isolate render target gating tests with fresh temp dirs
Tests in the "render target gating" describe block shared a single temp
directory via beforeAll/afterAll. The first test ran --only claude
(generating .claude/ files), and the second ran --only warp expecting no
Claude files — but leftovers from test 1 caused the assertion to fail.
Changing to beforeEach/afterEach gives each test a clean directory.
Closes #377
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(engine): add pre-sync commit guard and interactive apply mode
Adds two safety features to the sync pipeline:
1. Pre-sync commit guard: Detects uncommitted changes in protected
directories (.agentkit/engines, spec, overlays) before sync runs.
In TTY mode, prompts to abort, stash, or continue. In non-TTY
mode (CI), prints a warning and proceeds.
2. Interactive apply mode: After rendering, shows a change summary
and prompts: apply all / skip all / prompt each file. Per-file
prompt supports show-diff and apply-all-remaining. Default in
TTY; bypassed with --yes, --no-prompt, or --force.
New module: sync-guard.mjs with 4 exported functions and 7 tests.
New CLI flags: --yes, --no-prompt for non-interactive sync.
* chore(sync): regenerate outputs after branch merges
* chore(sync): regenerate outputs after merge of new-user-entry-point
* feat(engine): add configurable package manager and fix review findings
- Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn)
- Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.)
- Update CLAUDE.md template and hook/workflow templates to use {{packageManager}}
- Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch
- Remove duplicate cost-ops team definition from teams.yaml
- Pin all dependency versions in package.json (remove ^ prefixes)
- Add vitest coverage thresholds (80% statements/branches/functions/lines)
- Fix src/start/ code quality: null guards, exit delay, error boundaries
- Harden consolidate-branches.sh: self-resolution guard, stash restore
- Regenerate all 533 output files via agentkit sync
* chore(engine): add brand color palette variables to sync vars
* fix(ci): fix test race condition, workspace config, and lockfile
- Fix ConversationFlow test: wait for 'Got it' before asserting hint text
- Add packages field to pnpm-workspace.yaml for proper workspace resolution
- Regenerate lockfile after version pinning (removed ^ prefixes)
- Remove accidental .agentkit/templates/src/ artifacts
* fix(start): add event loop yields to ConversationFlow tests
ink-select-input needs setImmediate yields after a new menu renders
before it can process ENTER keypresses. Without this, the second ENTER
in leaf-selection tests was swallowed, causing intermittent failures.
---------
* fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391)
* Add entry point for new framework users (#389)
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* test(start): add coverage for result screen exit path
Add App-level test that navigates through ConversationFlow to a leaf
node and verifies the result screen text is rendered before exit().
Confirms the synchronous useEffect exit is safe — React commits the
render (Ink captures the frame) before useEffect fires.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: update generated sync output timestamps
Generated files updated with current sync date from dev merge.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
---------
* fix: caldues heuristics (#398)
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(quality): resolve all lint and format errors
- Fix Prettier formatting across engine src and start components
- Resolve 381 markdown lint errors (MD040, MD024, MD036, MD029, MD022, MD001)
- Add markdownlint config rules for intentional doc patterns
- Add .claude/worktrees to prettierignore to exclude external branches
- Enable markdownlint MD024/MD026 for duplicate headings and trailing colons
All quality checks now pass: Prettier (0 errors), Markdown lint (0 errors), Tests (111/111), Build (28.9kb)
* I'll update the last_updated field in all the files from '2026-03-12' to '2026-03-13'. This appears to be a routine timestamp update across all the AgentKit Forge generated files.
* docs: add AgentKit Forge sync feedback
Add detailed feedback on AgentKit Forge v3.1.0 integration, focusing on:
- Windows line-ending issues and multi-editor sync behavior
- Documentation scaffold-once limitations and override challenges
- Unresolved placeholder warnings lacking diagnostics
- Windows-specific pnpm execution problems
* docs: update CLAUDE.md with repository-specific editing guidelines
Clarified that modifications to `.agentkit` files are permitted in the agentkit-forge repository, while upstream directories remain protected. This ensures users understand the editing boundaries for project configuration and template files.
* docs(claude): allow .agentkit edits in this repo (NB for framework dev)
Made-with: Cursor
* chore(sync): regenerate outputs after agentkit:sync
Made-with: Cursor
* Fix/generated files and conflict markers (#427)
* fix(infra): resolve container app fqdn attribute and format code
* chore(sync): update AGENT_BACKLOG.md and other files for task management
- Enhanced AGENT_BACKLOG.md with detailed task scopes for CI pipeline configuration and test framework setup.
- Added new docker-compose.yml for local/staging validation of the framework.
- Updated CONTRIBUTING.md to include documentation hub link in the Discovery phase.
- Introduced README.md files in db, infra, and migrations directories to clarify their purpose for adopters.
- Added API conventions documentation to guide adopters on structuring their APIs.
- Created implementation plan for state management improvements and added relevant tests.
- Regenerated outputs across various files to reflect recent changes and ensure consistency.
* chore(ci): reduce CodeQL to weekly + manual only (#430)
* chore(ci): reduce CodeQL to weekly schedule + manual trigger
Removes push and pull_request triggers to reduce GitHub Actions costs.
Scans were running on every PR including Renovate dependency updates.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Potential fix for pull request finding
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* docs(architecture): add tool-neutral agent hub findings, ADR-10, and adoption roadmap (#428)
* docs(architecture): add tool-neutral agent hub findings, ADR-10, and adoption roadmap
Comparative analysis of agentkit-forge sync engine vs Mystira.workspace
hand-authored .agents/ pattern. Documents 5-phase adoption roadmap to
converge both approaches: .agents/ as sync target, reflective guards,
.readme.yaml generation, cross-session traces, and schema formalisation.
Includes regenerated sync output (updated timestamps across all tools).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs(architecture): add competitive landscape and strategic research report
Comprehensive analysis of…
JustAGhosT
added a commit
that referenced
this pull request
Mar 21, 2026
* Add start command: new user entry point with state detection (#387)
* fix(commands): add AskUserQuestion to VALID_TOOLS and /start command
The /start command session got stuck because AskUserQuestion was not
included in the allowed-tools whitelist. This fix addresses three issues:
1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite,
Agent, and NotebookEdit — preventing commands from declaring these
Claude Code built-in tools in their allowed-tools.
2. The /start command template now explicitly includes AskUserQuestion
in its allowed-tools frontmatter and instructs the agent to use it
for interactive guided choices (Phase 3).
3. Added /start command spec to commands.yaml with AskUserQuestion as
a declared tool dependency.
Also adds a test case validating all four newly-added tools are accepted
by the spec validator.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
* fix(commands): remove unrendered Handlebars comment from start.md output
The generated .claude/commands/start.md contained a raw {{! ... }}
Handlebars comment that was not processed by the sync engine. Remove it
so the generated output is clean Markdown.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
---------
* Add configurable prefix to kits commands (#388)
* feat(sync): add configurable command prefix for generated slash commands
Add `commandPrefix` setting to overlay settings that namespaces all
generated slash commands across platforms:
- Claude Code: subdirectory strategy (kits/check.md → /project:kits:check)
- Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check)
- Team commands excluded from prefixing (already namespaced)
Changes:
- Add resolveCommandPath() helper with subdirectory/filename strategies
- Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands,
syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills
- Add commandPrefix to vars from overlay settings
- Add commandPrefixedName to buildCommandVars
- Add 16 unit + integration tests (all pass, 93 existing tests unaffected)
Default is null (no prefix) for full backwards compatibility.
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
* fix(sync): address review findings for command prefix
- Remove unused afterAll import from test file
- Add clarifying comment that non-spec command files are also prefixed
- Add 2 integration tests verifying commandPrefixedName template variable
renders correctly with and without prefix
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
---------
* fix(ci): CI remediation — package manager, review findings, test stability (#390)
* fix(ci): resolve 7 bugs from project review
- BUG-001: Replace flaky discover test with controlled temp fixture
- BUG-002: Run prettier --write to fix formatting drift
- BUG-003: Add form-template detection skip in issue label validation
- BUG-005: Change claude.yml to self-hosted runner
- BUG-006: Align branch protection required status checks with project.yaml
- BUG-007: Fix command injection in resolve-merge.sh (use grep -F)
* docs: update changelog, add planning registry review findings
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed)
- Add Project Review Findings section to planning registry (PR-001 to PR-014)
- Update planning docs after sync merge
* feat(review): add --generate-plans flag to project-review command
Add Phase 2.5 plan generation after project review findings. When
--generate-plans is passed (default: true), scaffold plan files from
critical/high findings into docs/planning/review-findings/.
Also includes sync cleanup of stale cursor/windsurf settings.
* chore(sync): regenerate all outputs after project review fixes
* feat(cli): dynamic flag loading from commands.yaml + context-aware template hook
- Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags()
that reads flag definitions from commands.yaml at startup
- CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml
- Self-validation warns at startup if any flag is missing a type definition
- Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency
- Fix scaffold-once orphan bug: carry forward manifest entries for files
skipped by scaffold-once so orphan cleanup does not delete them
- Make protect-templates hook context-aware: skip protection in the
agentkit-forge source repo (detected via package.json name) so
maintainer agents can edit templates; block only in downstream repos
* feat(sync): add managed-mode script templates for downstream repos
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
* feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130)
Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating,
validating, and deploying new agent team specifications. Adapted from
cogmesh #130 with a simplified 6-agent pipeline:
- input-clarifier: assess requests, extract constraints
- mission-definer: lock team definition (ID, scope, accepts)
- role-architect: design agent roles and dependencies
- prompt-engineer: write agent descriptions and rules
- flow-designer: design team command and integration points
- team-validator: quality gate for spec consistency
Includes /team-forge command with --task flag (create-team, validate-team,
audit-teams, update-team) and planning doc.
* feat(teams): add Strategic Ops team for cross-project coordination
Add the STRATEGIC OPS team (T12) — handles framework governance,
portfolio analysis, adoption strategy, impact assessment, and release
coordination across all repos using AgentKit Forge.
5-agent pipeline:
- portfolio-analyst: inventory repos, detect drift, adoption metrics
- governance-advisor: versioning strategy, breaking change protocols
- adoption-strategist: onboarding, migration paths, rollout plans
- impact-assessor: blast radius analysis for template/spec changes
- release-coordinator: version bumps, sync waves, release comms
Includes /team-strategic-ops command with --task and --scope flags.
* feat(agents): add agent/team relationship matrix analysis engine + scripts
Add comprehensive agent/team relationship analysis with 8 cross-reference
matrices and 10 supplementary analyses (orphans, cycles, bottlenecks,
reachability, critical path, notification amplifiers, etc.).
- Fix YAML structure: strategic-ops agents now under own top-level key
- Add explicit agents: lists to forge + strategic-ops teams in teams.yaml
- Add consolidation detection responsibilities to portfolio-analyst
- Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers)
- Wire analyze-agents CLI command with --output/--matrix/--format flags
- Add managed-scaffold script templates (bash + PowerShell)
- Integrate into sync pipeline (auto-regenerates matrix on spec changes)
- Add 33 tests covering all matrices, analyses, and edge cases
* chore: update documentation files and add plan template
- Add trailing newlines to Cursor command documentation files for consistency
- Add new plan template files for project planning
- Improve markdown table formatting in Claude skills documentation
- Remove obsolete .clinerules/testing.md file
- Update various rule files with better formatting and advisory rule alignment
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
* feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364)
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365)
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
---------
* fix(teams): address CodeRabbit review findings on strategic-ops PR
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent
crash when node_modules is missing
- Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1
templates to fix invalid PowerShell syntax
- Fix protect-templates hook path: use 3 parent traversals to reach
.agentkit/package.json from hooks directory
- Remove schema-invalid 'phase' field from ai-cost-ops rule domain
- Narrow strategic-ops scope: replace **/* catch-all with specific files
- Add strategic-ops to cost-ops handoff-chain for consistency
- Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst,
and grant-hunter agents
- Fix glob matching in resolve-merge.sh (use regex instead of broken
sed strip)
- Add merge failure vs conflict detection in resolve-merge.ps1
- Add branch existence check in setup-agentkit-branch-governance scripts
- Add gh CLI preflight check in sync-split-pr.ps1
- Deduplicate branch protection loop when defaultBranch is 'main'
- Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/**
- Fix non-canonical doc paths in intake-agent-proposal.md
- Add changelog entries for new teams, agents, and analysis engine
- Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope)
- Run prettier on all modified files
* fix(templates): add gh auth preflight and changelog divider handling
- sync-split-pr.ps1: add gh auth status check before side effects
- update-changelog.ps1: stop before --- divider when appending entries
Addresses CodeRabbit review comments #7 and #13 on PR #356.
* fix(sync): set executable permission on analyze-agents.sh
Linux CI sync produces +x permissions; align local to match.
* fix(templates): address CodeRabbit review round 3-4 findings
- Fix YAML frontmatter in copilot agent template: use double quotes
for description field to handle apostrophes (CRITICAL)
- Fix protect-templates.sh/ps1 path traversal: correct parent
directory count for .agentkit/package.json source repo detection
- Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1
so native command failures (git, pnpm, gh) are treated as fatal
- Replace try/catch with $LASTEXITCODE check for gh auth status
- Deduplicate branch loop in setup-agentkit-branch-governance.sh
when defaultBranch equals 'main'
- Fix duplicate verification echo lines in governance scripts
* style(docs): fix prettier formatting on planning documents
Run prettier --write on web-intake-expansion.md and
intake-agent-proposal.md to fix CI formatting check.
* fix(tests): isolate render target gating tests with fresh temp dirs
Tests in the "render target gating" describe block shared a single temp
directory via beforeAll/afterAll. The first test ran --only claude
(generating .claude/ files), and the second ran --only warp expecting no
Claude files — but leftovers from test 1 caused the assertion to fail.
Changing to beforeEach/afterEach gives each test a clean directory.
Closes #377
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(engine): add pre-sync commit guard and interactive apply mode
Adds two safety features to the sync pipeline:
1. Pre-sync commit guard: Detects uncommitted changes in protected
directories (.agentkit/engines, spec, overlays) before sync runs.
In TTY mode, prompts to abort, stash, or continue. In non-TTY
mode (CI), prints a warning and proceeds.
2. Interactive apply mode: After rendering, shows a change summary
and prompts: apply all / skip all / prompt each file. Per-file
prompt supports show-diff and apply-all-remaining. Default in
TTY; bypassed with --yes, --no-prompt, or --force.
New module: sync-guard.mjs with 4 exported functions and 7 tests.
New CLI flags: --yes, --no-prompt for non-interactive sync.
* chore(sync): regenerate outputs after branch merges
* chore(sync): regenerate outputs after merge of new-user-entry-point
* feat(engine): add configurable package manager and fix review findings
- Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn)
- Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.)
- Update CLAUDE.md template and hook/workflow templates to use {{packageManager}}
- Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch
- Remove duplicate cost-ops team definition from teams.yaml
- Pin all dependency versions in package.json (remove ^ prefixes)
- Add vitest coverage thresholds (80% statements/branches/functions/lines)
- Fix src/start/ code quality: null guards, exit delay, error boundaries
- Harden consolidate-branches.sh: self-resolution guard, stash restore
- Regenerate all 533 output files via agentkit sync
* chore(engine): add brand color palette variables to sync vars
* fix(ci): fix test race condition, workspace config, and lockfile
- Fix ConversationFlow test: wait for 'Got it' before asserting hint text
- Add packages field to pnpm-workspace.yaml for proper workspace resolution
- Regenerate lockfile after version pinning (removed ^ prefixes)
- Remove accidental .agentkit/templates/src/ artifacts
* fix(start): add event loop yields to ConversationFlow tests
ink-select-input needs setImmediate yields after a new menu renders
before it can process ENTER keypresses. Without this, the second ENTER
in leaf-selection tests was swallowed, causing intermittent failures.
---------
* fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391)
* Add entry point for new framework users (#389)
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* test(start): add coverage for result screen exit path
Add App-level test that navigates through ConversationFlow to a leaf
node and verifies the result screen text is rendered before exit().
Confirms the synchronous useEffect exit is safe — React commits the
render (Ink captures the frame) before useEffect fires.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: update generated sync output timestamps
Generated files updated with current sync date from dev merge.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
---------
* fix: caldues heuristics (#398)
* feat: complete revisit of agents (#399) (#400)
* Add start command: new user entry point with state detection (#387)
* fix(commands): add AskUserQuestion to VALID_TOOLS and /start command
The /start command session got stuck because AskUserQuestion was not
included in the allowed-tools whitelist. This fix addresses three issues:
1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite,
Agent, and NotebookEdit — preventing commands from declaring these
Claude Code built-in tools in their allowed-tools.
2. The /start command template now explicitly includes AskUserQuestion
in its allowed-tools frontmatter and instructs the agent to use it
for interactive guided choices (Phase 3).
3. Added /start command spec to commands.yaml with AskUserQuestion as
a declared tool dependency.
Also adds a test case validating all four newly-added tools are accepted
by the spec validator.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
* fix(commands): remove unrendered Handlebars comment from start.md output
The generated .claude/commands/start.md contained a raw {{! ... }}
Handlebars comment that was not processed by the sync engine. Remove it
so the generated output is clean Markdown.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
---------
* Add configurable prefix to kits commands (#388)
* feat(sync): add configurable command prefix for generated slash commands
Add `commandPrefix` setting to overlay settings that namespaces all
generated slash commands across platforms:
- Claude Code: subdirectory strategy (kits/check.md → /project:kits:check)
- Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check)
- Team commands excluded from prefixing (already namespaced)
Changes:
- Add resolveCommandPath() helper with subdirectory/filename strategies
- Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands,
syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills
- Add commandPrefix to vars from overlay settings
- Add commandPrefixedName to buildCommandVars
- Add 16 unit + integration tests (all pass, 93 existing tests unaffected)
Default is null (no prefix) for full backwards compatibility.
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
* fix(sync): address review findings for command prefix
- Remove unused afterAll import from test file
- Add clarifying comment that non-spec command files are also prefixed
- Add 2 integration tests verifying commandPrefixedName template variable
renders correctly with and without prefix
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
---------
* fix(ci): CI remediation — package manager, review findings, test stability (#390)
* fix(ci): resolve 7 bugs from project review
- BUG-001: Replace flaky discover test with controlled temp fixture
- BUG-002: Run prettier --write to fix formatting drift
- BUG-003: Add form-template detection skip in issue label validation
- BUG-005: Change claude.yml to self-hosted runner
- BUG-006: Align branch protection required status checks with project.yaml
- BUG-007: Fix command injection in resolve-merge.sh (use grep -F)
* docs: update changelog, add planning registry review findings
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed)
- Add Project Review Findings section to planning registry (PR-001 to PR-014)
- Update planning docs after sync merge
* feat(review): add --generate-plans flag to project-review command
Add Phase 2.5 plan generation after project review findings. When
--generate-plans is passed (default: true), scaffold plan files from
critical/high findings into docs/planning/review-findings/.
Also includes sync cleanup of stale cursor/windsurf settings.
* chore(sync): regenerate all outputs after project review fixes
* feat(cli): dynamic flag loading from commands.yaml + context-aware template hook
- Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags()
that reads flag definitions from commands.yaml at startup
- CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml
- Self-validation warns at startup if any flag is missing a type definition
- Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency
- Fix scaffold-once orphan bug: carry forward manifest entries for files
skipped by scaffold-once so orphan cleanup does not delete them
- Make protect-templates hook context-aware: skip protection in the
agentkit-forge source repo (detected via package.json name) so
maintainer agents can edit templates; block only in downstream repos
* feat(sync): add managed-mode script templates for downstream repos
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
* feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130)
Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating,
validating, and deploying new agent team specifications. Adapted from
cogmesh #130 with a simplified 6-agent pipeline:
- input-clarifier: assess requests, extract constraints
- mission-definer: lock team definition (ID, scope, accepts)
- role-architect: design agent roles and dependencies
- prompt-engineer: write agent descriptions and rules
- flow-designer: design team command and integration points
- team-validator: quality gate for spec consistency
Includes /team-forge command with --task flag (create-team, validate-team,
audit-teams, update-team) and planning doc.
* feat(teams): add Strategic Ops team for cross-project coordination
Add the STRATEGIC OPS team (T12) — handles framework governance,
portfolio analysis, adoption strategy, impact assessment, and release
coordination across all repos using AgentKit Forge.
5-agent pipeline:
- portfolio-analyst: inventory repos, detect drift, adoption metrics
- governance-advisor: versioning strategy, breaking change protocols
- adoption-strategist: onboarding, migration paths, rollout plans
- impact-assessor: blast radius analysis for template/spec changes
- release-coordinator: version bumps, sync waves, release comms
Includes /team-strategic-ops command with --task and --scope flags.
* feat(agents): add agent/team relationship matrix analysis engine + scripts
Add comprehensive agent/team relationship analysis with 8 cross-reference
matrices and 10 supplementary analyses (orphans, cycles, bottlenecks,
reachability, critical path, notification amplifiers, etc.).
- Fix YAML structure: strategic-ops agents now under own top-level key
- Add explicit agents: lists to forge + strategic-ops teams in teams.yaml
- Add consolidation detection responsibilities to portfolio-analyst
- Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers)
- Wire analyze-agents CLI command with --output/--matrix/--format flags
- Add managed-scaffold script templates (bash + PowerShell)
- Integrate into sync pipeline (auto-regenerates matrix on spec changes)
- Add 33 tests covering all matrices, analyses, and edge cases
* chore: update documentation files and add plan template
- Add trailing newlines to Cursor command documentation files for consistency
- Add new plan template files for project planning
- Improve markdown table formatting in Claude skills documentation
- Remove obsolete .clinerules/testing.md file
- Update various rule files with better formatting and advisory rule alignment
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
* feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364)
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365)
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
---------
* fix(teams): address CodeRabbit review findings on strategic-ops PR
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent
crash when node_modules is missing
- Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1
templates to fix invalid PowerShell syntax
- Fix protect-templates hook path: use 3 parent traversals to reach
.agentkit/package.json from hooks directory
- Remove schema-invalid 'phase' field from ai-cost-ops rule domain
- Narrow strategic-ops scope: replace **/* catch-all with specific files
- Add strategic-ops to cost-ops handoff-chain for consistency
- Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst,
and grant-hunter agents
- Fix glob matching in resolve-merge.sh (use regex instead of broken
sed strip)
- Add merge failure vs conflict detection in resolve-merge.ps1
- Add branch existence check in setup-agentkit-branch-governance scripts
- Add gh CLI preflight check in sync-split-pr.ps1
- Deduplicate branch protection loop when defaultBranch is 'main'
- Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/**
- Fix non-canonical doc paths in intake-agent-proposal.md
- Add changelog entries for new teams, agents, and analysis engine
- Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope)
- Run prettier on all modified files
* fix(templates): add gh auth preflight and changelog divider handling
- sync-split-pr.ps1: add gh auth status check before side effects
- update-changelog.ps1: stop before --- divider when appending entries
Addresses CodeRabbit review comments #7 and #13 on PR #356.
* fix(sync): set executable permission on analyze-agents.sh
Linux CI sync produces +x permissions; align local to match.
* fix(templates): address CodeRabbit review round 3-4 findings
- Fix YAML frontmatter in copilot agent template: use double quotes
for description field to handle apostrophes (CRITICAL)
- Fix protect-templates.sh/ps1 path traversal: correct parent
directory count for .agentkit/package.json source repo detection
- Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1
so native command failures (git, pnpm, gh) are treated as fatal
- Replace try/catch with $LASTEXITCODE check for gh auth status
- Deduplicate branch loop in setup-agentkit-branch-governance.sh
when defaultBranch equals 'main'
- Fix duplicate verification echo lines in governance scripts
* style(docs): fix prettier formatting on planning documents
Run prettier --write on web-intake-expansion.md and
intake-agent-proposal.md to fix CI formatting check.
* fix(tests): isolate render target gating tests with fresh temp dirs
Tests in the "render target gating" describe block shared a single temp
directory via beforeAll/afterAll. The first test ran --only claude
(generating .claude/ files), and the second ran --only warp expecting no
Claude files — but leftovers from test 1 caused the assertion to fail.
Changing to beforeEach/afterEach gives each test a clean directory.
Closes #377
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(engine): add pre-sync commit guard and interactive apply mode
Adds two safety features to the sync pipeline:
1. Pre-sync commit guard: Detects uncommitted changes in protected
directories (.agentkit/engines, spec, overlays) before sync runs.
In TTY mode, prompts to abort, stash, or continue. In non-TTY
mode (CI), prints a warning and proceeds.
2. Interactive apply mode: After rendering, shows a change summary
and prompts: apply all / skip all / prompt each file. Per-file
prompt supports show-diff and apply-all-remaining. Default in
TTY; bypassed with --yes, --no-prompt, or --force.
New module: sync-guard.mjs with 4 exported functions and 7 tests.
New CLI flags: --yes, --no-prompt for non-interactive sync.
* chore(sync): regenerate outputs after branch merges
* chore(sync): regenerate outputs after merge of new-user-entry-point
* feat(engine): add configurable package manager and fix review findings
- Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn)
- Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.)
- Update CLAUDE.md template and hook/workflow templates to use {{packageManager}}
- Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch
- Remove duplicate cost-ops team definition from teams.yaml
- Pin all dependency versions in package.json (remove ^ prefixes)
- Add vitest coverage thresholds (80% statements/branches/functions/lines)
- Fix src/start/ code quality: null guards, exit delay, error boundaries
- Harden consolidate-branches.sh: self-resolution guard, stash restore
- Regenerate all 533 output files via agentkit sync
* chore(engine): add brand color palette variables to sync vars
* fix(ci): fix test race condition, workspace config, and lockfile
- Fix ConversationFlow test: wait for 'Got it' before asserting hint text
- Add packages field to pnpm-workspace.yaml for proper workspace resolution
- Regenerate lockfile after version pinning (removed ^ prefixes)
- Remove accidental .agentkit/templates/src/ artifacts
* fix(start): add event loop yields to ConversationFlow tests
ink-select-input needs setImmediate yields after a new menu renders
before it can process ENTER keypresses. Without this, the second ENTER
in leaf-selection tests was swallowed, causing intermittent failures.
---------
* fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391)
* Add entry point for new framework users (#389)
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* test(start): add coverage for result screen exit path
Add App-level test that navigates through ConversationFlow to a leaf
node and verifies the result screen text is rendered before exit().
Confirms the synchronous useEffect exit is safe — React commits the
render (Ink captures the frame) before useEffect fires.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: update generated sync output timestamps
Generated files updated with current sync date from dev merge.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
---------
* fix: caldues heuristics (#398)
---------
* fix(quality): resolve all lint and format errors
- Fix Prettier formatting across engine src and start components
- Resolve 381 markdown lint errors (MD040, MD024, MD036, MD029, MD022, MD001)
- Add markdownlint config rules for intentional doc patterns
- Add .claude/worktrees to prettierignore to exclude external branches
- Enable markdownlint MD024/MD026 for duplicate headings and trailing colons
All quality checks now pass: Prettier (0 errors), Markdown lint (0 errors), Tests (111/111), Build (28.9kb)
* I'll update the last_updated field in all the files from '2026-03-12' to '2026-03-13'. This appears to be a routine timestamp update across all the AgentKit Forge generated files.
* docs: add AgentKit Forge sync feedback
Add detailed feedback on AgentKit Forge v3.1.0 integration, focusing on:
- Windows line-ending issues and multi-editor sync behavior
- Documentation scaffold-once limitations and override challenges
- Unresolved placeholder warnings lacking diagnostics
- Windows-specific pnpm execution problems
* docs: update CLAUDE.md with repository-specific editing guidelines
Clarified that modifications to `.agentkit` files are permitted in the agentkit-forge repository, while upstream directories remain protected. This ensures users understand the editing boundaries for project configuration and template files.
* docs(claude): allow .agentkit edits in this repo (NB for framework dev)
Made-with: Cursor
* chore(sync): regenerate outputs after agentkit:sync
Made-with: Cursor
* Fix/generated files and conflict markers (#427)
* fix(infra): resolve container app fqdn attribute and format code
* chore(sync): update AGENT_BACKLOG.md and other files for task management
- Enhanced AGENT_BACKLOG.md with detailed task scopes for CI pipeline configuration and test framework setup.
- Added new docker-compose.yml for local/staging validation of the framework.
- Updated CONTRIBUTING.md to include documentation hub link in the Discovery phase.
- Introduced README.md files in db, infra, and migrations directories to clarify their purpose for adopters.
- Added API conventions documentation to guide adopters on structuring their APIs.
- Created implementation plan for state management improvements and added relevant tests.
- Regenerated outputs across various files to reflect recent changes and ensure consistency.
* chore(ci): reduce CodeQL to weekly + manual only (#430)
* chore(ci): reduce CodeQL to weekly schedule + manual trigger
Removes push and pull_request triggers to reduce GitHub Actions costs.
Scans were running on every PR including Renovate dependency updates.
* Potential fix for pull request finding
---------
* docs(architecture): add tool-neutral agent hub findings, ADR-10, and adoption roadmap (#428)
* docs(architecture): add tool-neutral agent hub findings, ADR-10, and adoption roadmap
Comparative analysis of agentkit-forge sync engine vs Mystira.workspace
hand-authored .agents/ pattern. Documents 5-phase adoption roadmap to
converge both approaches: .agents/ as sync target, reflective guards,
.readme.yaml generation, cross-session traces, and schema formalisation.
Includes regenerated sync output (updated timestamps across all tools).
* docs(architecture): add competitive landscape and strategic research report
Comprehensive analysis of…
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
JustAGhosT
added a commit
that referenced
this pull request
Mar 28, 2026
… scaffold policy enhancements
* Add start command: new user entry point with state detection (#387)
* fix(commands): add AskUserQuestion to VALID_TOOLS and /start command
The /start command session got stuck because AskUserQuestion was not
included in the allowed-tools whitelist. This fix addresses three issues:
1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite,
Agent, and NotebookEdit — preventing commands from declaring these
Claude Code built-in tools in their allowed-tools.
2. The /start command template now explicitly includes AskUserQuestion
in its allowed-tools frontmatter and instructs the agent to use it
for interactive guided choices (Phase 3).
3. Added /start command spec to commands.yaml with AskUserQuestion as
a declared tool dependency.
Also adds a test case validating all four newly-added tools are accepted
by the spec validator.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
* fix(commands): remove unrendered Handlebars comment from start.md output
The generated .claude/commands/start.md contained a raw {{! ... }}
Handlebars comment that was not processed by the sync engine. Remove it
so the generated output is clean Markdown.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
---------
Co-authored-by: Claude <noreply@anthropic.com>
* Add configurable prefix to kits commands (#388)
* feat(sync): add configurable command prefix for generated slash commands
Add `commandPrefix` setting to overlay settings that namespaces all
generated slash commands across platforms:
- Claude Code: subdirectory strategy (kits/check.md → /project:kits:check)
- Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check)
- Team commands excluded from prefixing (already namespaced)
Changes:
- Add resolveCommandPath() helper with subdirectory/filename strategies
- Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands,
syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills
- Add commandPrefix to vars from overlay settings
- Add commandPrefixedName to buildCommandVars
- Add 16 unit + integration tests (all pass, 93 existing tests unaffected)
Default is null (no prefix) for full backwards compatibility.
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
* fix(sync): address review findings for command prefix
- Remove unused afterAll import from test file
- Add clarifying comment that non-spec command files are also prefixed
- Add 2 integration tests verifying commandPrefixedName template variable
renders correctly with and without prefix
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(ci): CI remediation — package manager, review findings, test stability (#390)
* fix(ci): resolve 7 bugs from project review
- BUG-001: Replace flaky discover test with controlled temp fixture
- BUG-002: Run prettier --write to fix formatting drift
- BUG-003: Add form-template detection skip in issue label validation
- BUG-005: Change claude.yml to self-hosted runner
- BUG-006: Align branch protection required status checks with project.yaml
- BUG-007: Fix command injection in resolve-merge.sh (use grep -F)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: update changelog, add planning registry review findings
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed)
- Add Project Review Findings section to planning registry (PR-001 to PR-014)
- Update planning docs after sync merge
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(review): add --generate-plans flag to project-review command
Add Phase 2.5 plan generation after project review findings. When
--generate-plans is passed (default: true), scaffold plan files from
critical/high findings into docs/planning/review-findings/.
Also includes sync cleanup of stale cursor/windsurf settings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate all outputs after project review fixes
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cli): dynamic flag loading from commands.yaml + context-aware template hook
- Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags()
that reads flag definitions from commands.yaml at startup
- CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml
- Self-validation warns at startup if any flag is missing a type definition
- Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency
- Fix scaffold-once orphan bug: carry forward manifest entries for files
skipped by scaffold-once so orphan cleanup does not delete them
- Make protect-templates hook context-aware: skip protection in the
agentkit-forge source repo (detected via package.json name) so
maintainer agents can edit templates; block only in downstream repos
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(sync): add managed-mode script templates for downstream repos
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130)
Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating,
validating, and deploying new agent team specifications. Adapted from
cogmesh #130 with a simplified 6-agent pipeline:
- input-clarifier: assess requests, extract constraints
- mission-definer: lock team definition (ID, scope, accepts)
- role-architect: design agent roles and dependencies
- prompt-engineer: write agent descriptions and rules
- flow-designer: design team command and integration points
- team-validator: quality gate for spec consistency
Includes /team-forge command with --task flag (create-team, validate-team,
audit-teams, update-team) and planning doc.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(teams): add Strategic Ops team for cross-project coordination
Add the STRATEGIC OPS team (T12) — handles framework governance,
portfolio analysis, adoption strategy, impact assessment, and release
coordination across all repos using AgentKit Forge.
5-agent pipeline:
- portfolio-analyst: inventory repos, detect drift, adoption metrics
- governance-advisor: versioning strategy, breaking change protocols
- adoption-strategist: onboarding, migration paths, rollout plans
- impact-assessor: blast radius analysis for template/spec changes
- release-coordinator: version bumps, sync waves, release comms
Includes /team-strategic-ops command with --task and --scope flags.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(agents): add agent/team relationship matrix analysis engine + scripts
Add comprehensive agent/team relationship analysis with 8 cross-reference
matrices and 10 supplementary analyses (orphans, cycles, bottlenecks,
reachability, critical path, notification amplifiers, etc.).
- Fix YAML structure: strategic-ops agents now under own top-level key
- Add explicit agents: lists to forge + strategic-ops teams in teams.yaml
- Add consolidation detection responsibilities to portfolio-analyst
- Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers)
- Wire analyze-agents CLI command with --output/--matrix/--format flags
- Add managed-scaffold script templates (bash + PowerShell)
- Integrate into sync pipeline (auto-regenerates matrix on spec changes)
- Add 33 tests covering all matrices, analyses, and edge cases
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: update documentation files and add plan template
- Add trailing newlines to Cursor command documentation files for consistency
- Add new plan template files for project planning
- Improve markdown table formatting in Claude skills documentation
- Remove obsolete .clinerules/testing.md file
- Update various rule files with better formatting and advisory rule alignment
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364)
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365)
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(teams): address CodeRabbit review findings on strategic-ops PR
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent
crash when node_modules is missing
- Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1
templates to fix invalid PowerShell syntax
- Fix protect-templates hook path: use 3 parent traversals to reach
.agentkit/package.json from hooks directory
- Remove schema-invalid 'phase' field from ai-cost-ops rule domain
- Narrow strategic-ops scope: replace **/* catch-all with specific files
- Add strategic-ops to cost-ops handoff-chain for consistency
- Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst,
and grant-hunter agents
- Fix glob matching in resolve-merge.sh (use regex instead of broken
sed strip)
- Add merge failure vs conflict detection in resolve-merge.ps1
- Add branch existence check in setup-agentkit-branch-governance scripts
- Add gh CLI preflight check in sync-split-pr.ps1
- Deduplicate branch protection loop when defaultBranch is 'main'
- Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/**
- Fix non-canonical doc paths in intake-agent-proposal.md
- Add changelog entries for new teams, agents, and analysis engine
- Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope)
- Run prettier on all modified files
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): add gh auth preflight and changelog divider handling
- sync-split-pr.ps1: add gh auth status check before side effects
- update-changelog.ps1: stop before --- divider when appending entries
Addresses CodeRabbit review comments #7 and #13 on PR #356.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(sync): set executable permission on analyze-agents.sh
Linux CI sync produces +x permissions; align local to match.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(templates): address CodeRabbit review round 3-4 findings
- Fix YAML frontmatter in copilot agent template: use double quotes
for description field to handle apostrophes (CRITICAL)
- Fix protect-templates.sh/ps1 path traversal: correct parent
directory count for .agentkit/package.json source repo detection
- Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1
so native command failures (git, pnpm, gh) are treated as fatal
- Replace try/catch with $LASTEXITCODE check for gh auth status
- Deduplicate branch loop in setup-agentkit-branch-governance.sh
when defaultBranch equals 'main'
- Fix duplicate verification echo lines in governance scripts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* style(docs): fix prettier formatting on planning documents
Run prettier --write on web-intake-expansion.md and
intake-agent-proposal.md to fix CI formatting check.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tests): isolate render target gating tests with fresh temp dirs
Tests in the "render target gating" describe block shared a single temp
directory via beforeAll/afterAll. The first test ran --only claude
(generating .claude/ files), and the second ran --only warp expecting no
Claude files — but leftovers from test 1 caused the assertion to fail.
Changing to beforeEach/afterEach gives each test a clean directory.
Closes #377
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(engine): add pre-sync commit guard and interactive apply mode
Adds two safety features to the sync pipeline:
1. Pre-sync commit guard: Detects uncommitted changes in protected
directories (.agentkit/engines, spec, overlays) before sync runs.
In TTY mode, prompts to abort, stash, or continue. In non-TTY
mode (CI), prints a warning and proceeds.
2. Interactive apply mode: After rendering, shows a change summary
and prompts: apply all / skip all / prompt each file. Per-file
prompt supports show-diff and apply-all-remaining. Default in
TTY; bypassed with --yes, --no-prompt, or --force.
New module: sync-guard.mjs with 4 exported functions and 7 tests.
New CLI flags: --yes, --no-prompt for non-interactive sync.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate outputs after branch merges
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(sync): regenerate outputs after merge of new-user-entry-point
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(engine): add configurable package manager and fix review findings
- Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn)
- Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.)
- Update CLAUDE.md template and hook/workflow templates to use {{packageManager}}
- Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch
- Remove duplicate cost-ops team definition from teams.yaml
- Pin all dependency versions in package.json (remove ^ prefixes)
- Add vitest coverage thresholds (80% statements/branches/functions/lines)
- Fix src/start/ code quality: null guards, exit delay, error boundaries
- Harden consolidate-branches.sh: self-resolution guard, stash restore
- Regenerate all 533 output files via agentkit sync
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore(engine): add brand color palette variables to sync vars
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(ci): fix test race condition, workspace config, and lockfile
- Fix ConversationFlow test: wait for 'Got it' before asserting hint text
- Add packages field to pnpm-workspace.yaml for proper workspace resolution
- Regenerate lockfile after version pinning (removed ^ prefixes)
- Remove accidental .agentkit/templates/src/ artifacts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(start): add event loop yields to ConversationFlow tests
ink-select-input needs setImmediate yields after a new menu renders
before it can process ENTER keypresses. Without this, the second ENTER
in leaf-selection tests was swallowed, causing intermittent failures.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391)
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Add entry point for new framework users (#389)
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* test(start): add coverage for result screen exit path
Add App-level test that navigates through ConversationFlow to a leaf
node and verifies the result screen text is rendered before exit().
Confirms the synchronous useEffect exit is safe — React commits the
render (Ink captures the frame) before useEffect fires.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: update generated sync output timestamps
Generated files updated with current sync date from dev merge.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix: caldues heuristics (#398)
* feat: complete revisit of agents (#399) (#400)
* Add start command: new user entry point with state detection (#387)
* fix(commands): add AskUserQuestion to VALID_TOOLS and /start command
The /start command session got stuck because AskUserQuestion was not
included in the allowed-tools whitelist. This fix addresses three issues:
1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite,
Agent, and NotebookEdit — preventing commands from declaring these
Claude Code built-in tools in their allowed-tools.
2. The /start command template now explicitly includes AskUserQuestion
in its allowed-tools frontmatter and instructs the agent to use it
for interactive guided choices (Phase 3).
3. Added /start command spec to commands.yaml with AskUserQuestion as
a declared tool dependency.
Also adds a test case validating all four newly-added tools are accepted
by the spec validator.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
* fix(commands): remove unrendered Handlebars comment from start.md output
The generated .claude/commands/start.md contained a raw {{! ... }}
Handlebars comment that was not processed by the sync engine. Remove it
so the generated output is clean Markdown.
https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M
---------
* Add configurable prefix to kits commands (#388)
* feat(sync): add configurable command prefix for generated slash commands
Add `commandPrefix` setting to overlay settings that namespaces all
generated slash commands across platforms:
- Claude Code: subdirectory strategy (kits/check.md → /project:kits:check)
- Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check)
- Team commands excluded from prefixing (already namespaced)
Changes:
- Add resolveCommandPath() helper with subdirectory/filename strategies
- Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands,
syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills
- Add commandPrefix to vars from overlay settings
- Add commandPrefixedName to buildCommandVars
- Add 16 unit + integration tests (all pass, 93 existing tests unaffected)
Default is null (no prefix) for full backwards compatibility.
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
* fix(sync): address review findings for command prefix
- Remove unused afterAll import from test file
- Add clarifying comment that non-spec command files are also prefixed
- Add 2 integration tests verifying commandPrefixedName template variable
renders correctly with and without prefix
https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p
---------
* fix(ci): CI remediation — package manager, review findings, test stability (#390)
* fix(ci): resolve 7 bugs from project review
- BUG-001: Replace flaky discover test with controlled temp fixture
- BUG-002: Run prettier --write to fix formatting drift
- BUG-003: Add form-template detection skip in issue label validation
- BUG-005: Change claude.yml to self-hosted runner
- BUG-006: Align branch protection required status checks with project.yaml
- BUG-007: Fix command injection in resolve-merge.sh (use grep -F)
* docs: update changelog, add planning registry review findings
- Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed)
- Add Project Review Findings section to planning registry (PR-001 to PR-014)
- Update planning docs after sync merge
* feat(review): add --generate-plans flag to project-review command
Add Phase 2.5 plan generation after project review findings. When
--generate-plans is passed (default: true), scaffold plan files from
critical/high findings into docs/planning/review-findings/.
Also includes sync cleanup of stale cursor/windsurf settings.
* chore(sync): regenerate all outputs after project review fixes
* feat(cli): dynamic flag loading from commands.yaml + context-aware template hook
- Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags()
that reads flag definitions from commands.yaml at startup
- CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml
- Self-validation warns at startup if any flag is missing a type definition
- Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency
- Fix scaffold-once orphan bug: carry forward manifest entries for files
skipped by scaffold-once so orphan cleanup does not delete them
- Make protect-templates hook context-aware: skip protection in the
agentkit-forge source repo (detected via package.json name) so
maintainer agents can edit templates; block only in downstream repos
* feat(sync): add managed-mode script templates for downstream repos
Add 14 script templates (.agentkit/templates/scripts/) with `managed`
scaffold mode so downstream repos receive script updates via three-way
merge while preserving local customizations.
Templates include: create-doc, update-changelog, validate-documentation,
validate-numbering, check-documentation-requirement, sync-issues,
sync-split-pr, setup-agentkit-branch-governance, and resolve-merge
(both .sh and .ps1 variants where applicable).
Parameterized templates use {{defaultBranch}} and branch protection
variables from project.yaml. Engine wired via syncScripts() under
doc-scaffolding feature gate.
* feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130)
Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating,
validating, and deploying new agent team specifications. Adapted from
cogmesh #130 with a simplified 6-agent pipeline:
- input-clarifier: assess requests, extract constraints
- mission-definer: lock team definition (ID, scope, accepts)
- role-architect: design agent roles and dependencies
- prompt-engineer: write agent descriptions and rules
- flow-designer: design team command and integration points
- team-validator: quality gate for spec consistency
Includes /team-forge command with --task flag (create-team, validate-team,
audit-teams, update-team) and planning doc.
* feat(teams): add Strategic Ops team for cross-project coordination
Add the STRATEGIC OPS team (T12) — handles framework governance,
portfolio analysis, adoption strategy, impact assessment, and release
coordination across all repos using AgentKit Forge.
5-agent pipeline:
- portfolio-analyst: inventory repos, detect drift, adoption metrics
- governance-advisor: versioning strategy, breaking change protocols
- adoption-strategist: onboarding, migration paths, rollout plans
- impact-assessor: blast radius analysis for template/spec changes
- release-coordinator: version bumps, sync waves, release comms
Includes /team-strategic-ops command with --task and --scope flags.
* feat(agents): add agent/team relationship matrix analysis engine + scripts
Add comprehensive agent/team relationship analysis with 8 cross-reference
matrices and 10 supplementary analyses (orphans, cycles, bottlenecks,
reachability, critical path, notification amplifiers, etc.).
- Fix YAML structure: strategic-ops agents now under own top-level key
- Add explicit agents: lists to forge + strategic-ops teams in teams.yaml
- Add consolidation detection responsibilities to portfolio-analyst
- Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers)
- Wire analyze-agents CLI command with --output/--matrix/--format flags
- Add managed-scaffold script templates (bash + PowerShell)
- Integrate into sync pipeline (auto-regenerates matrix on spec changes)
- Add 33 tests covering all matrices, analyses, and edge cases
* chore: update documentation files and add plan template
- Add trailing newlines to Cursor command documentation files for consistency
- Add new plan template files for project planning
- Improve markdown table formatting in Claude skills documentation
- Remove obsolete .clinerules/testing.md file
- Update various rule files with better formatting and advisory rule alignment
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
* feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364)
* feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14)
- Add spec-compliance-auditor to operations category (quality team) — closes
the feedback loop between agent specs and actual behavior
- Add Cost Ops team (T14) with 5 agents: model-economist,
token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter,
cost-ops-monitor
- Add ai-cost-ops rules domain with 6 conventions (model routing, token
budgets, caching, batch APIs, vendor abstraction, credit tracking)
- Add team-cost-ops command with --task, --period, --provider flags
- Update notification chains: data→cost-ops-monitor, infra→model-economist,
retrospective-analyst→spec-compliance-auditor
- Add intake routes: cost-ops, agent-performance
* fix(teams): add implement to cost-ops team accepts list
Resolves agent/team accepts mismatch — token-efficiency-engineer accepts
implement but the team definition only had investigate/review/plan/document.
* feat(cost-ops): add multi-provider infra cost ticket to backlog
Add detailed planning ticket for multi-provider infrastructure cost
normalisation, routing, and cost-agent integration. Covers 9 providers
(Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS,
GCP) with 3-phase delivery plan and cross-team dependency tracking.
* fix(templates): resolve PR review comments from CodeRabbit
- Fix protect-templates.sh: package name check uses correct
"agentkit-forge-runtime" instead of "agentkit-forge"
- Fix protect-templates.ps1: malformed path (missing separator before
.agentkit) and same package name correction
- Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS
require() syntax used in the inline Node script
- Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder
- Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left
by sync engine, keeping user formatting + implement in cost-ops
* feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365)
Enable web research capabilities for cost-ops team:
- Add WebSearch and WebFetch to /team-cost-ops allowed-tools
- Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst,
and grant-hunter agent preferred-tools
- Add web intake expansion ticket (P2) for MCP crawler, Puppeteer
integration, automated pricing refresh, and cross-session persistence
- Regenerate sync outputs
---------
* fix(teams): address CodeRabbit review findings on strategic-ops PR
- Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent
crash when node_modules is missing
- Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1
templates to fix invalid PowerShell syntax
- Fix protect-templates hook path: use 3 parent traversals to reach
.agentkit/package.json from hooks directory
- Remove schema-invalid 'phase' field from ai-cost-ops rule domain
- Narrow strategic-ops scope: replace **/* catch-all with specific files
- Add strategic-ops to cost-ops handoff-chain for consistency
- Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst,
and grant-hunter agents
- Fix glob matching in resolve-merge.sh (use regex instead of broken
sed strip)
- Add merge failure vs conflict detection in resolve-merge.ps1
- Add branch existence check in setup-agentkit-branch-governance scripts
- Add gh CLI preflight check in sync-split-pr.ps1
- Deduplicate branch protection loop when defaultBranch is 'main'
- Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/**
- Fix non-canonical doc paths in intake-agent-proposal.md
- Add changelog entries for new teams, agents, and analysis engine
- Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope)
- Run prettier on all modified files
* fix(templates): add gh auth preflight and changelog divider handling
- sync-split-pr.ps1: add gh auth status check before side effects
- update-changelog.ps1: stop before --- divider when appending entries
Addresses CodeRabbit review comments #7 and #13 on PR #356.
* fix(sync): set executable permission on analyze-agents.sh
Linux CI sync produces +x permissions; align local to match.
* fix(templates): address CodeRabbit review round 3-4 findings
- Fix YAML frontmatter in copilot agent template: use double quotes
for description field to handle apostrophes (CRITICAL)
- Fix protect-templates.sh/ps1 path traversal: correct parent
directory count for .agentkit/package.json source repo detection
- Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1
so native command failures (git, pnpm, gh) are treated as fatal
- Replace try/catch with $LASTEXITCODE check for gh auth status
- Deduplicate branch loop in setup-agentkit-branch-governance.sh
when defaultBranch equals 'main'
- Fix duplicate verification echo lines in governance scripts
* style(docs): fix prettier formatting on planning documents
Run prettier --write on web-intake-expansion.md and
intake-agent-proposal.md to fix CI formatting check.
* fix(tests): isolate render target gating tests with fresh temp dirs
Tests in the "render target gating" describe block shared a single temp
directory via beforeAll/afterAll. The first test ran --only claude
(generating .claude/ files), and the second ran --only warp expecting no
Claude files — but leftovers from test 1 caused the assertion to fail.
Changing to beforeEach/afterEach gives each test a clean directory.
Closes #377
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(engine): add pre-sync commit guard and interactive apply mode
Adds two safety features to the sync pipeline:
1. Pre-sync commit guard: Detects uncommitted changes in protected
directories (.agentkit/engines, spec, overlays) before sync runs.
In TTY mode, prompts to abort, stash, or continue. In non-TTY
mode (CI), prints a warning and proceeds.
2. Interactive apply mode: After rendering, shows a change summary
and prompts: apply all / skip all / prompt each file. Per-file
prompt supports show-diff and apply-all-remaining. Default in
TTY; bypassed with --yes, --no-prompt, or --force.
New module: sync-guard.mjs with 4 exported functions and 7 tests.
New CLI flags: --yes, --no-prompt for non-interactive sync.
* chore(sync): regenerate outputs after branch merges
* chore(sync): regenerate outputs after merge of new-user-entry-point
* feat(engine): add configurable package manager and fix review findings
- Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn)
- Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.)
- Update CLAUDE.md template and hook/workflow templates to use {{packageManager}}
- Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch
- Remove duplicate cost-ops team definition from teams.yaml
- Pin all dependency versions in package.json (remove ^ prefixes)
- Add vitest coverage thresholds (80% statements/branches/functions/lines)
- Fix src/start/ code quality: null guards, exit delay, error boundaries
- Harden consolidate-branches.sh: self-resolution guard, stash restore
- Regenerate all 533 output files via agentkit sync
* chore(engine): add brand color palette variables to sync vars
* fix(ci): fix test race condition, workspace config, and lockfile
- Fix ConversationFlow test: wait for 'Got it' before asserting hint text
- Add packages field to pnpm-workspace.yaml for proper workspace resolution
- Regenerate lockfile after version pinning (removed ^ prefixes)
- Remove accidental .agentkit/templates/src/ artifacts
* fix(start): add event loop yields to ConversationFlow tests
ink-select-input needs setImmediate yields after a new menu renders
before it can process ENTER keypresses. Without this, the second ENTER
in leaf-selection tests was swallowed, causing intermittent failures.
---------
* fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391)
* Add entry point for new framework users (#389)
* feat(commands): add /start command as new user entry point
Adds a context-aware triage command that detects repository state
(fresh clone, post-discovery, mid-session, uncommitted work) and
guides users to the right command or team for their goal.
Includes team routing table, 4 contextual flows, and decision
guidance for when users don't know which team to use.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* refactor(commands): add Arguments and State Management sections to /start
Addresses TEAMFORGE validation findings:
- Add dedicated Arguments section documenting $ARGUMENTS handling
- Add State Management section (reads/writes inventory)
- Fix frontmatter: remove misleading generated_by field
- Add explicit "manually authored" comment header
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): add /start to spec and template for cross-repo generation
Moves /start from a hand-authored command to a proper spec-driven,
sync-generated command available to any repo that adopts AgentKit Forge.
- Add start command definition to commands.yaml (no feature gate — always on)
- Create start.md template in .agentkit/templates/claude/commands/
- Add /start to CLAUDE.md Quick Reference table template
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after adding /start command
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(commands): make /start team routing dynamic instead of hardcoded
Replace the static team routing table with dynamic discovery:
1. Read AGENT_TEAMS.md (from /discover) at runtime
2. Fall back to .agentkit/spec/teams.yaml
3. Fall back to /team-* command frontmatter
This ensures /start always reflects the actual teams configured
in any repo, rather than a hardcoded list that could go stale.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore(sync): regenerate outputs after dynamic routing change
Generated by: pnpm -C .agentkit agentkit:sync
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add interactive TUI entry point with hybrid UI
Replace static markdown output with an ink-based terminal UI that
combines two modes: a guided conversation flow for first-run users
and a fuzzy-searchable command palette for returning users. A
persistent status bar shows repo state at a glance (branch, phase,
backlog count, working tree status).
- Context detection module mirrors /start Phase 1 signals
- ConversationFlow: branching dialogue tree (choose-your-own-adventure)
- CommandPalette: fuzzy search with context-ranked star recommendations
- StatusBar: tmux-style persistent strip with color-coded segments
- Supports --json flag for scripting/piping
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* feat(start): add test suite, error handling, and refactor for production readiness
- Add vitest test infrastructure with 102 tests across 7 files
- Coverage: 95.58% statements, 90.81% branches, 96.72% functions
- Extract conversation tree to separate config module for testability
- Add error boundary to App component for graceful error display
- Add --help flag with usage documentation
- Add TTY detection with JSON fallback for non-interactive environments
- Add SIGINT/SIGTERM signal handling for clean exit
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: add coverage to gitignore, include plan and workspace config
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — unused code, git cwd, magic numbers
- Call exit() after command selection so Ink process terminates
- Remove unused ctx prop from Header component
- Fix git commands to use -C flag with root path parameter
- Extract dumpContextJson() helper to deduplicate JSON output logic
- Add comment explaining hardcoded team filter exclusion
- Replace magic numbers with named constants (FUSE_THRESHOLD,
RECOMMENDED_SCORE, MAX_BRANCH_LENGTH)
- Remove unused ink-spinner dependency
- Add test verifying git -C flag passes root correctly
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* fix(start): address review findings — security, correctness, test quality
- CRITICAL: Replace execSync shell interpolation with execFileSync to
prevent command injection via root parameter in detect.js
- CRITICAL: Remove phantom --external:ink-spinner from build script and
switch npx to direct esbuild invocation
- HIGH: Fix exit race condition — use useEffect instead of setTimeout
- HIGH: Wire up onSelect callback in ConversationFlow so Guide mode
triggers the result screen in App
- HIGH: Destructure and accept ctx prop in ConversationFlow
- HIGH: Add back-navigation (Escape) in ConversationFlow
- HIGH: Guard against undefined team.focus in commands.js tags
- MEDIUM: Use functional setCursor form to avoid stale closures
- MEDIUM: Key commandIndices Map by string id instead of object identity
- MEDIUM: Improve parseTeams header detection (drop first row approach)
- MEDIUM: Filter completed/done/closed items from backlogCount
- MEDIUM: Extract shared makeCtx test utility across all test files
- MEDIUM: Add null guard to StatusBar truncate helper
- LOW: Validate orchestratorPhase is a number in range 1-5
- LOW: Fix pnpm-workspace.yaml list syntax
- Replace all setTimeout in tests with vi.waitFor deterministic waits
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* test(start): add coverage for result screen exit path
Add App-level test that navigates through ConversationFlow to a leaf
node and verifies the result screen text is rendered before exit().
Confirms the synchronous useEffect exit is safe — React commits the
render (Ink captures the frame) before useEffect fires.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
* chore: update generated sync output timestamps
Generated files updated with current sync date from dev merge.
https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW
---------
* fix: caldues heuristics (#398)
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(quality): resolve all lint and format errors
- Fix Prettier formatting across engine src and start components
- Resolve 381 markdown lint errors (MD040, MD024, MD036, MD029, MD022, MD001)
- Add markdownlint config rules for intentional doc patterns
- Add .claude/worktrees to prettierignore to exclude external branches
- Enable markdownlint MD024/MD026 for duplicate headings and trailing colons
All quality checks now pass: Prettier (0 errors), Markdown lint (0 errors), Tests (111/111), Build (28.9kb)
* I'll update the last_updated field in all the files from '2026-03-12' to '2026-03-13'. This appears to be a routine timestamp update across all the AgentKit Forge generated files.
* docs: add AgentKit Forge sync feedback
Add detailed feedback on AgentKit Forge v3.1.0 integration, focusing on:
- Windows line-ending issues and multi-editor sync behavior
- Documentation scaffold-once limitations and override challenges
- Unresolved placeholder warnings lacking diagnostics
- Windows-specific pnpm execution problems
* docs: update CLAUDE.md with repository-specific editing guidelines
Clarified that modifications to `.agentkit` files are permitted in the agentkit-forge repository, while upstream directories remain protected. This ensures users understand the editing boundaries for project configuration and template files.
* docs(claude): allow .agentkit edits in this repo (NB for framework dev)
Made-with: Cursor
* chore(sync): regenerate outputs after agentkit:sync
Made-with: Cursor
* Fix/generated files and conflict markers (#427)
* fix(infra): resolve container app fqdn attribute and format code
* chore(sync): update AGENT_BACKLOG.md and other files for task management
- Enhanced AGENT_BACKLOG.md with detailed task scopes for CI pipeline configuration and test framework setup.
- Added new docker-compose.yml for local/staging validation of the framework.
- Updated CONTRIBUTING.md to include documentation hub link in the Discovery phase.
- Introduced README.md files in db, infra, and migrations directories to clarify their purpose for adopters.
- Added API conventions documentation to guide adopters on structuring their APIs.
- Created implementation plan for state management improvements and added relevant tests.
- Regenerated outputs across various files to reflect recent changes and ensure consistency.
* chore(ci): reduce CodeQL to weekly + manual only (#430)
* chore(ci): reduce CodeQL to weekly schedule + manual trigger
Removes push and pull_request triggers to reduce GitHub Actions costs.
Scans were running on every PR including Renovate dependency updates.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Potential fix for pull request finding
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* docs(architecture): add tool-neutral agent hub findings, ADR-10, and adoption roadmap (#428)
* docs(architecture): add tool-neutral agent hub findings, ADR-10, and adoption roadmap
Comparative analysis of agentkit-forge sync engine vs Mystira.workspace
hand-authored .agents/ pattern. Documents 5-phase adoption roadmap to
converge both approaches: .agents/ as sync target, reflective guards,
.readme.yaml generation, cross-session traces, and schema formalisation.
Includes regenerated sync output (updated timestamps across all tools).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs(architecture): add competi…
JustAGhosT
added a commit
that referenced
this pull request
Mar 28, 2026
… scaffold policy enhancements (#474) * Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills - Add commandPrefix to vars from overlay settings - Add commandPrefixedName to buildCommandVars - Add 16 unit + integration tests (all pass, 93 existing tests unaffected) Default is null (no prefix) for full backwards compatibility. https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p * fix(sync): address review findings for command prefix - Remove unused afterAll import from test file - Add clarifying comment that non-spec command files are also prefixed - Add 2 integration tests verifying commandPrefixedName template variable renders correctly with and without prefix https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p --------- * fix(ci): CI remediation — package manager, review findings, test stability (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. * chore(sync): regenerate all outputs after project review fixes * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs --------- * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. * chore(sync): regenerate outputs after branch merges * chore(sync): regenerate outputs after merge of new-user-entry-point * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync * chore(engine): add brand color palette variables to sync vars * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. --------- * fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391) * Add entry point for new framework users (#389) * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * test(start): add coverage for result screen exit path Add App-level test that navigates through ConversationFlow to a leaf node and verifies the result screen text is rendered before exit(). Confirms the synchronous useEffect exit is safe — React commits the render (Ink captures the frame) before useEffect fires. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: update generated sync output timestamps Generated files updated with current sync date from dev merge. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW --------- * fix: caldues heuristics (#398) * feat: complete revisit of agents (#399) (#400) * Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills - Add commandPrefix to vars from overlay settings - Add commandPrefixedName to buildCommandVars - Add 16 unit + integration tests (all pass, 93 existing tests unaffected) Default is null (no prefix) for full backwards compatibility. https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p * fix(sync): address review findings for command prefix - Remove unused afterAll import from test file - Add clarifying comment that non-spec command files are also prefixed - Add 2 integration tests verifying commandPrefixedName template variable renders correctly with and without prefix https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p --------- * fix(ci): CI remediation — package manager, review findings, test stability (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. * chore(sync): regenerate all outputs after project review fixes * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs --------- * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. * chore(sync): regenerate outputs after branch merges * chore(sync): regenerate outputs after merge of new-user-entry-point * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync * chore(engine): add brand color palette variables to sync vars * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. --------- * fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391) * Add entry point for new framework users (#389) * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * test(start): add coverage for result screen exit path Add App-level test that navigates through ConversationFlow to a leaf node and verifies the result screen text is rendered before exit(). Confirms the synchronous useEffect exit is safe — React commits the render (Ink captures the frame) before useEffect fires. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: update generated sync output timestamps Generated files updated with current sync date from dev merge. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW --------- * fix: caldues heuristics (#398) --------- * fix(quality): resolve all lint and format errors - Fix Prettier formatting across engine src and start components - Resolve 381 markdown lint errors (MD040, MD024, MD036, MD029, MD022, MD001) - Add markdownlint config rules for intentional doc patterns - Add .claude/worktrees to prettierignore to exclude external branches - Enable markdownlint MD024/MD026 for duplicate headings and trailing colons All quality checks now pass: Prettier (0 errors), Markdown lint (0 errors), Tests (111/111), Build (28.9kb) * I'll update the last_updated field in all the files from '2026-03-12' to '2026-03-13'. This appears to be a routine timestamp update across all the AgentKit Forge generated files. * docs: add AgentKit Forge sync feedback Add detailed feedback on AgentKit Forge v3.1.0 integration, focusing on: - Windows line-ending issues and multi-editor sync behavior - Documentation scaffold-once limitations and override challenges - Unresolved placeholder warnings lacking diagnostics - Windows-specific pnpm execution problems * docs: update CLAUDE.md with repository-specific editing guidelines Clarified that modifications to `.agentkit` files are permitted in the agentkit-forge repository, while upstream directories remain protected. This ensures users understand the editing boundaries for project configuration and template files. * docs(claude): allow .agentkit edits in this repo (NB for framework dev) Made-with: Cursor * chore(sync): regenerate outputs after agentkit:sync Made-with: Cursor * Fix/generated files and conflict markers (#427) * fix(infra): resolve container app fqdn attribute and format code * chore(sync): update AGENT_BACKLOG.md and other files for task management - Enhanced AGENT_BACKLOG.md with detailed task scopes for CI pipeline configuration and test framework setup. - Added new docker-compose.yml for local/staging validation of the framework. - Updated CONTRIBUTING.md to include documentation hub link in the Discovery phase. - Introduced README.md files in db, infra, and migrations directories to clarify their purpose for adopters. - Added API conventions documentation to guide adopters on structuring their APIs. - Created implementation plan for state management improvements and added relevant tests. - Regenerated outputs across various files to reflect recent changes and ensure consistency. * chore(ci): reduce CodeQL to weekly + manual only (#430) * chore(ci): reduce CodeQL to weekly schedule + manual trigger Removes push and pull_request triggers to reduce GitHub Actions costs. Scans were running on every PR including Renovate dependency updates. * Potential fix for pull request finding --------- * docs(architecture): add tool-neutral agent hub findings, ADR-10, and adoption roadmap (#428) * docs(architecture): add tool-neutral agent hub findings, ADR-10, and adoption roadmap Comparative analysis of agentkit-forge sync engine vs Mystira.workspace hand-authored .agents/ pattern. Documents 5-phase adoption roadmap to converge both approaches: .agents/ as sync target, reflective guards, .readme.yaml generation, cross-session traces, and schema formalisation. Includes regenerated sync output (updated timestamps across all tools). * docs(architecture): add competi… Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com> Co-authored-by: CodeRabbit <noreply@coderabbit.ai>
JustAGhosT
added a commit
that referenced
this pull request
Apr 1, 2026
…nores (#507) * Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- Co-authored-by: Claude <noreply@anthropic.com> * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills - Add commandPrefix to vars from overlay settings - Add commandPrefixedName to buildCommandVars - Add 16 unit + integration tests (all pass, 93 existing tests unaffected) Default is null (no prefix) for full backwards compatibility. https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p * fix(sync): address review findings for command prefix - Remove unused afterAll import from test file - Add clarifying comment that non-spec command files are also prefixed - Add 2 integration tests verifying commandPrefixedName template variable renders correctly with and without prefix https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p --------- Co-authored-by: Claude <noreply@anthropic.com> * fix(ci): CI remediation — package manager, review findings, test stability (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate all outputs after project review fixes Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate outputs after branch merges Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(sync): regenerate outputs after merge of new-user-entry-point Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore(engine): add brand color palette variables to sync vars Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391) Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> * Add entry point for new framework users (#389) * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * test(start): add coverage for result screen exit path Add App-level test that navigates through ConversationFlow to a leaf node and verifies the result screen text is rendered before exit(). Confirms the synchronous useEffect exit is safe — React commits the render (Ink captures the frame) before useEffect fires. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: update generated sync output timestamps Generated files updated with current sync date from dev merge. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW --------- Co-authored-by: Claude <noreply@anthropic.com> * fix: caldues heuristics (#398) * feat: complete revisit of agents (#399) (#400) * Add start command: new user entry point with state detection (#387) * fix(commands): add AskUserQuestion to VALID_TOOLS and /start command The /start command session got stuck because AskUserQuestion was not included in the allowed-tools whitelist. This fix addresses three issues: 1. VALID_TOOLS in spec-validator was missing AskUserQuestion, TodoWrite, Agent, and NotebookEdit — preventing commands from declaring these Claude Code built-in tools in their allowed-tools. 2. The /start command template now explicitly includes AskUserQuestion in its allowed-tools frontmatter and instructs the agent to use it for interactive guided choices (Phase 3). 3. Added /start command spec to commands.yaml with AskUserQuestion as a declared tool dependency. Also adds a test case validating all four newly-added tools are accepted by the spec validator. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M * fix(commands): remove unrendered Handlebars comment from start.md output The generated .claude/commands/start.md contained a raw {{! ... }} Handlebars comment that was not processed by the sync engine. Remove it so the generated output is clean Markdown. https://claude.ai/code/session_01Qh3Xk3jFkVdeRAXTqLvq3M --------- * Add configurable prefix to kits commands (#388) * feat(sync): add configurable command prefix for generated slash commands Add `commandPrefix` setting to overlay settings that namespaces all generated slash commands across platforms: - Claude Code: subdirectory strategy (kits/check.md → /project:kits:check) - Cursor/Windsurf/Copilot/Codex: filename prefix (kits-check.md → /kits-check) - Team commands excluded from prefixing (already namespaced) Changes: - Add resolveCommandPath() helper with subdirectory/filename strategies - Update syncClaudeCommands, syncClaudeSkills, syncCursorCommands, syncWindsurfCommands, syncCopilotPrompts, syncCodexSkills - Add commandPrefix to vars from overlay settings - Add commandPrefixedName to buildCommandVars - Add 16 unit + integration tests (all pass, 93 existing tests unaffected) Default is null (no prefix) for full backwards compatibility. https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p * fix(sync): address review findings for command prefix - Remove unused afterAll import from test file - Add clarifying comment that non-spec command files are also prefixed - Add 2 integration tests verifying commandPrefixedName template variable renders correctly with and without prefix https://claude.ai/code/session_01EBjmVEhi7fP2huAL3SBR6p --------- * fix(ci): CI remediation — package manager, review findings, test stability (#390) * fix(ci): resolve 7 bugs from project review - BUG-001: Replace flaky discover test with controlled temp fixture - BUG-002: Run prettier --write to fix formatting drift - BUG-003: Add form-template detection skip in issue label validation - BUG-005: Change claude.yml to self-hosted runner - BUG-006: Align branch protection required status checks with project.yaml - BUG-007: Fix command injection in resolve-merge.sh (use grep -F) * docs: update changelog, add planning registry review findings - Add changelog entries for Wave 1-3 fixes (Added/Changed/Fixed/Removed) - Add Project Review Findings section to planning registry (PR-001 to PR-014) - Update planning docs after sync merge * feat(review): add --generate-plans flag to project-review command Add Phase 2.5 plan generation after project review findings. When --generate-plans is passed (default: true), scaffold plan files from critical/high findings into docs/planning/review-findings/. Also includes sync cleanup of stale cursor/windsurf settings. * chore(sync): regenerate all outputs after project review fixes * feat(cli): dynamic flag loading from commands.yaml + context-aware template hook - Replace ~200 lines of hardcoded VALID_FLAGS/FLAG_TYPES with loadCommandFlags() that reads flag definitions from commands.yaml at startup - CLI_INTERNAL_FLAGS/CLI_INTERNAL_FLAG_TYPES cover commands not in commands.yaml - Self-validation warns at startup if any flag is missing a type definition - Update cli.test.mjs to validate CLI_INTERNAL_FLAGS consistency - Fix scaffold-once orphan bug: carry forward manifest entries for files skipped by scaffold-once so orphan cleanup does not delete them - Make protect-templates hook context-aware: skip protection in the agentkit-forge source repo (detected via package.json name) so maintainer agents can edit templates; block only in downstream repos * feat(sync): add managed-mode script templates for downstream repos Add 14 script templates (.agentkit/templates/scripts/) with `managed` scaffold mode so downstream repos receive script updates via three-way merge while preserving local customizations. Templates include: create-doc, update-changelog, validate-documentation, validate-numbering, check-documentation-requirement, sync-issues, sync-split-pr, setup-agentkit-branch-governance, and resolve-merge (both .sh and .ps1 variants where applicable). Parameterized templates use {{defaultBranch}} and branch protection variables from project.yaml. Engine wired via syncScripts() under doc-scaffolding feature gate. * feat(teams): add TeamForge meta-team for agent team creation (cogmesh #130) Add the TEAMFORGE meta-team (T11) — a structured pipeline for creating, validating, and deploying new agent team specifications. Adapted from cogmesh #130 with a simplified 6-agent pipeline: - input-clarifier: assess requests, extract constraints - mission-definer: lock team definition (ID, scope, accepts) - role-architect: design agent roles and dependencies - prompt-engineer: write agent descriptions and rules - flow-designer: design team command and integration points - team-validator: quality gate for spec consistency Includes /team-forge command with --task flag (create-team, validate-team, audit-teams, update-team) and planning doc. * feat(teams): add Strategic Ops team for cross-project coordination Add the STRATEGIC OPS team (T12) — handles framework governance, portfolio analysis, adoption strategy, impact assessment, and release coordination across all repos using AgentKit Forge. 5-agent pipeline: - portfolio-analyst: inventory repos, detect drift, adoption metrics - governance-advisor: versioning strategy, breaking change protocols - adoption-strategist: onboarding, migration paths, rollout plans - impact-assessor: blast radius analysis for template/spec changes - release-coordinator: version bumps, sync waves, release comms Includes /team-strategic-ops command with --task and --scope flags. * feat(agents): add agent/team relationship matrix analysis engine + scripts Add comprehensive agent/team relationship analysis with 8 cross-reference matrices and 10 supplementary analyses (orphans, cycles, bottlenecks, reachability, critical path, notification amplifiers, etc.). - Fix YAML structure: strategic-ops agents now under own top-level key - Add explicit agents: lists to forge + strategic-ops teams in teams.yaml - Add consolidation detection responsibilities to portfolio-analyst - Create agent-analysis.mjs engine module (loadFullAgentGraph + renderers) - Wire analyze-agents CLI command with --output/--matrix/--format flags - Add managed-scaffold script templates (bash + PowerShell) - Integrate into sync pipeline (auto-regenerates matrix on spec changes) - Add 33 tests covering all matrices, analyses, and edge cases * chore: update documentation files and add plan template - Add trailing newlines to Cursor command documentation files for consistency - Add new plan template files for project planning - Improve markdown table formatting in Claude skills documentation - Remove obsolete .clinerules/testing.md file - Update various rule files with better formatting and advisory rule alignment * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs * feat(cost-ops): add Cost Ops team, agents, and multi-provider cost ticket (#364) * feat(agents): add spec-compliance-auditor feedback loop + Cost Ops team (T14) - Add spec-compliance-auditor to operations category (quality team) — closes the feedback loop between agent specs and actual behavior - Add Cost Ops team (T14) with 5 agents: model-economist, token-efficiency-engineer, vendor-arbitrage-analyst, grant-hunter, cost-ops-monitor - Add ai-cost-ops rules domain with 6 conventions (model routing, token budgets, caching, batch APIs, vendor abstraction, credit tracking) - Add team-cost-ops command with --task, --period, --provider flags - Update notification chains: data→cost-ops-monitor, infra→model-economist, retrospective-analyst→spec-compliance-auditor - Add intake routes: cost-ops, agent-performance * fix(teams): add implement to cost-ops team accepts list Resolves agent/team accepts mismatch — token-efficiency-engineer accepts implement but the team definition only had investigate/review/plan/document. * feat(cost-ops): add multi-provider infra cost ticket to backlog Add detailed planning ticket for multi-provider infrastructure cost normalisation, routing, and cost-agent integration. Covers 9 providers (Anthropic, OpenAI, Google, Mistral, Cohere, self-hosted, Azure, AWS, GCP) with 3-phase delivery plan and cross-team dependency tracking. * fix(templates): resolve PR review comments from CodeRabbit - Fix protect-templates.sh: package name check uses correct "agentkit-forge-runtime" instead of "agentkit-forge" - Fix protect-templates.ps1: malformed path (missing separator before .agentkit) and same package name correction - Fix update-changelog.ps1: change .mjs to .cjs to match CommonJS require() syntax used in the inline Node script - Fix resolve-merge.sh: add fallback for {{defaultBranch}} placeholder - Fix AGENT_TEAMS.md: resolve three-way merge conflict markers left by sync engine, keeping user formatting + implement in cost-ops * feat(cost-ops): add WebSearch/WebFetch tools to cost-ops agents (#365) Enable web research capabilities for cost-ops team: - Add WebSearch and WebFetch to /team-cost-ops allowed-tools - Add WebSearch and WebFetch to model-economist, vendor-arbitrage-analyst, and grant-hunter agent preferred-tools - Add web intake expansion ticket (P2) for MCP crawler, Puppeteer integration, automated pricing refresh, and cross-session persistence - Regenerate sync outputs --------- * fix(teams): address CodeRabbit review findings on strategic-ops PR - Lazy-load js-yaml in cli.mjs after ensureDependencies() to prevent crash when node_modules is missing - Convert YAML frontmatter to PowerShell comment blocks in all 6 PS1 templates to fix invalid PowerShell syntax - Fix protect-templates hook path: use 3 parent traversals to reach .agentkit/package.json from hooks directory - Remove schema-invalid 'phase' field from ai-cost-ops rule domain - Narrow strategic-ops scope: replace **/* catch-all with specific files - Add strategic-ops to cost-ops handoff-chain for consistency - Add Write tool to spec-compliance-auditor, vendor-arbitrage-analyst, and grant-hunter agents - Fix glob matching in resolve-merge.sh (use regex instead of broken sed strip) - Add merge failure vs conflict detection in resolve-merge.ps1 - Add branch existence check in setup-agentkit-branch-governance scripts - Add gh CLI preflight check in sync-split-pr.ps1 - Deduplicate branch protection loop when defaultBranch is 'main' - Fix applies-to glob: docs/planning/cost/** → docs/planning/cost-governance/** - Fix non-canonical doc paths in intake-agent-proposal.md - Add changelog entries for new teams, agents, and analysis engine - Resolve AGENT_TEAMS.md merge conflict (accept narrowed scope) - Run prettier on all modified files * fix(templates): add gh auth preflight and changelog divider handling - sync-split-pr.ps1: add gh auth status check before side effects - update-changelog.ps1: stop before --- divider when appending entries Addresses CodeRabbit review comments #7 and #13 on PR #356. * fix(sync): set executable permission on analyze-agents.sh Linux CI sync produces +x permissions; align local to match. * fix(templates): address CodeRabbit review round 3-4 findings - Fix YAML frontmatter in copilot agent template: use double quotes for description field to handle apostrophes (CRITICAL) - Fix protect-templates.sh/ps1 path traversal: correct parent directory count for .agentkit/package.json source repo detection - Add $PSNativeCommandUseErrorActionPreference to sync-split-pr.ps1 so native command failures (git, pnpm, gh) are treated as fatal - Replace try/catch with $LASTEXITCODE check for gh auth status - Deduplicate branch loop in setup-agentkit-branch-governance.sh when defaultBranch equals 'main' - Fix duplicate verification echo lines in governance scripts * style(docs): fix prettier formatting on planning documents Run prettier --write on web-intake-expansion.md and intake-agent-proposal.md to fix CI formatting check. * fix(tests): isolate render target gating tests with fresh temp dirs Tests in the "render target gating" describe block shared a single temp directory via beforeAll/afterAll. The first test ran --only claude (generating .claude/ files), and the second ran --only warp expecting no Claude files — but leftovers from test 1 caused the assertion to fail. Changing to beforeEach/afterEach gives each test a clean directory. Closes #377 * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(engine): add pre-sync commit guard and interactive apply mode Adds two safety features to the sync pipeline: 1. Pre-sync commit guard: Detects uncommitted changes in protected directories (.agentkit/engines, spec, overlays) before sync runs. In TTY mode, prompts to abort, stash, or continue. In non-TTY mode (CI), prints a warning and proceeds. 2. Interactive apply mode: After rendering, shows a change summary and prompts: apply all / skip all / prompt each file. Per-file prompt supports show-diff and apply-all-remaining. Default in TTY; bypassed with --yes, --no-prompt, or --force. New module: sync-guard.mjs with 4 exported functions and 7 tests. New CLI flags: --yes, --no-prompt for non-interactive sync. * chore(sync): regenerate outputs after branch merges * chore(sync): regenerate outputs after merge of new-user-entry-point * feat(engine): add configurable package manager and fix review findings - Add `stack.packageManager` field to project.yaml (pnpm | npm | yarn) - Derive helper template vars (pmInstall, pmRun, pmExec, pmLockfile, etc.) - Update CLAUDE.md template and hook/workflow templates to use {{packageManager}} - Fix sync-guard to exclude .agentkit/spec from protected dirs and add try/catch - Remove duplicate cost-ops team definition from teams.yaml - Pin all dependency versions in package.json (remove ^ prefixes) - Add vitest coverage thresholds (80% statements/branches/functions/lines) - Fix src/start/ code quality: null guards, exit delay, error boundaries - Harden consolidate-branches.sh: self-resolution guard, stash restore - Regenerate all 533 output files via agentkit sync * chore(engine): add brand color palette variables to sync vars * fix(ci): fix test race condition, workspace config, and lockfile - Fix ConversationFlow test: wait for 'Got it' before asserting hint text - Add packages field to pnpm-workspace.yaml for proper workspace resolution - Regenerate lockfile after version pinning (removed ^ prefixes) - Remove accidental .agentkit/templates/src/ artifacts * fix(start): add event loop yields to ConversationFlow tests ink-select-input needs setImmediate yields after a new menu renders before it can process ENTER keypresses. Without this, the second ENTER in leaf-selection tests was swallowed, causing intermittent failures. --------- * fix(spec): correct githubSlug to phoenixvc/agentkit-forge (#391) * Add entry point for new framework users (#389) * feat(commands): add /start command as new user entry point Adds a context-aware triage command that detects repository state (fresh clone, post-discovery, mid-session, uncommitted work) and guides users to the right command or team for their goal. Includes team routing table, 4 contextual flows, and decision guidance for when users don't know which team to use. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * refactor(commands): add Arguments and State Management sections to /start Addresses TEAMFORGE validation findings: - Add dedicated Arguments section documenting $ARGUMENTS handling - Add State Management section (reads/writes inventory) - Fix frontmatter: remove misleading generated_by field - Add explicit "manually authored" comment header https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): add /start to spec and template for cross-repo generation Moves /start from a hand-authored command to a proper spec-driven, sync-generated command available to any repo that adopts AgentKit Forge. - Add start command definition to commands.yaml (no feature gate — always on) - Create start.md template in .agentkit/templates/claude/commands/ - Add /start to CLAUDE.md Quick Reference table template https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after adding /start command Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(commands): make /start team routing dynamic instead of hardcoded Replace the static team routing table with dynamic discovery: 1. Read AGENT_TEAMS.md (from /discover) at runtime 2. Fall back to .agentkit/spec/teams.yaml 3. Fall back to /team-* command frontmatter This ensures /start always reflects the actual teams configured in any repo, rather than a hardcoded list that could go stale. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore(sync): regenerate outputs after dynamic routing change Generated by: pnpm -C .agentkit agentkit:sync https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add interactive TUI entry point with hybrid UI Replace static markdown output with an ink-based terminal UI that combines two modes: a guided conversation flow for first-run users and a fuzzy-searchable command palette for returning users. A persistent status bar shows repo state at a glance (branch, phase, backlog count, working tree status). - Context detection module mirrors /start Phase 1 signals - ConversationFlow: branching dialogue tree (choose-your-own-adventure) - CommandPalette: fuzzy search with context-ranked star recommendations - StatusBar: tmux-style persistent strip with color-coded segments - Supports --json flag for scripting/piping https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * feat(start): add test suite, error handling, and refactor for production readiness - Add vitest test infrastructure with 102 tests across 7 files - Coverage: 95.58% statements, 90.81% branches, 96.72% functions - Extract conversation tree to separate config module for testability - Add error boundary to App component for graceful error display - Add --help flag with usage documentation - Add TTY detection with JSON fallback for non-interactive environments - Add SIGINT/SIGTERM signal handling for clean exit https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: add coverage to gitignore, include plan and workspace config https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — unused code, git cwd, magic numbers - Call exit() after command selection so Ink process terminates - Remove unused ctx prop from Header component - Fix git commands to use -C flag with root path parameter - Extract dumpContextJson() helper to deduplicate JSON output logic - Add comment explaining hardcoded team filter exclusion - Replace magic numbers with named constants (FUSE_THRESHOLD, RECOMMENDED_SCORE, MAX_BRANCH_LENGTH) - Remove unused ink-spinner dependency - Add test verifying git -C flag passes root correctly https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * fix(start): address review findings — security, correctness, test quality - CRITICAL: Replace execSync shell interpolation with execFileSync to prevent command injection via root parameter in detect.js - CRITICAL: Remove phantom --external:ink-spinner from build script and switch npx to direct esbuild invocation - HIGH: Fix exit race condition — use useEffect instead of setTimeout - HIGH: Wire up onSelect callback in ConversationFlow so Guide mode triggers the result screen in App - HIGH: Destructure and accept ctx prop in ConversationFlow - HIGH: Add back-navigation (Escape) in ConversationFlow - HIGH: Guard against undefined team.focus in commands.js tags - MEDIUM: Use functional setCursor form to avoid stale closures - MEDIUM: Key commandIndices Map by string id instead of object identity - MEDIUM: Improve parseTeams header detection (drop first row approach) - MEDIUM: Filter completed/done/closed items from backlogCount - MEDIUM: Extract shared makeCtx test utility across all test files - MEDIUM: Add null guard to StatusBar truncate helper - LOW: Validate orchestratorPhase is a number in range 1-5 - LOW: Fix pnpm-workspace.yaml list syntax - Replace all setTimeout in tests with vi.waitFor deterministic waits https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * test(start): add coverage for result screen exit path Add App-level test that navigates through ConversationFlow to a leaf node and verifies the result screen text is rendered before exit(). Confirms the synchronous useEffect exit is safe — React commits the render (Ink captures the frame) before useEffect fires. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW * chore: update generated sync output timestamps Generated files updated with current sync date from dev merge. https://claude.ai/code/session_01Gy1HvvywVe6Z2iFJ2bHYCW --------- * fix: caldues heuristics (#398) --------- Co-authored-by: Claude <noreply@anthropic.com> * fix(quality): resolve all lint and format errors - Fix Prettier formatting across engine src and start components - Resolve 381 markdown lint errors (MD040, MD024, MD036, MD029, MD022, MD001) - Add markdownlint config rules for intentional doc patterns - Add .claude/worktrees to prettierignore to exclude external branches - Enable markdownlint MD024/MD026 for duplicate headings and trailing colons All quality checks now pass: Prettier (0 errors), Markdown lint (0 errors), Tests (111/111), Build (28.9kb) * I'll update the last_updated field in all the files from '2026-03-12' to '2026-03-13'. This appears to be a routine timestamp update across all the AgentKit Forge generated files. * docs: add AgentKit Forge sync feedback Add detailed feedback on AgentKit Forge v3.1.0 integration, focusing on: - Windows line-ending issues and multi-editor sync behavior - Documentation scaffold-once limitations and override challenges - Unresolved placeholder warnings lacking diagnostics - Windows-specific pnpm execution problems * docs: update CLAUDE.md with repository-specific editing guidelines Clarified that modifications to `.agentkit` files are permitted in the agentkit-forge repository, while upstream directories remain protected. This ensures users understand the editing boundaries for project configuration and template files. * docs(claude): allow .agentkit edits in this repo (NB for framework dev) Made-with: Cursor * chore(sync): regenerate outputs after agentkit:sync Made-with: Cursor * Fix/generated files and conflict markers (#427) * fix(infra): resolve container app fqdn attribute and format code * chore(sync): update AGENT_BACKLOG.md and other files for task management - Enhanced AGENT_BACKLOG.md with detailed task scopes for CI pipeline configuration and test framework setup. - Added new docker-compose.yml for local/staging validation of the framework. - Updated CONTRIBUTING.md to include documentation hub link in the Discovery phase. - Introduced README.md files in db, infra, and migrations directories to clarify their purpose for adopters. - Added API conventions documentation to guide adopters on structuring their APIs. - Created implementation plan for state management improvements and added relevant tests. - Regenerated outputs across various files to reflect recent changes and ensure consistency. * chore(ci): reduce CodeQL to weekly + manual only (#430) * chore(ci): reduce CodeQL to weekly schedule + manual trigger Removes push and pull_request triggers to reduce GitHub Actions costs. Scans were running on every PR including Renovate dependency updates. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Potential fix for pull request finding Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com> * docs(architecture): add tool-neutral agent hub findings, ADR-10, and adoption roadmap (#428) * docs(architecture): add tool-neutral agent hub findings, ADR-10, and adoption roadmap Comparative analysis of agentkit-forge sync engine vs Mystira.workspace hand-authored .agents/ pattern. Documents 5-phase adoption roadmap to converge both approaches: .agents/ as sync target, reflective guards, .readme.yaml generation, cross-session traces, and schema formalisation. Includes regenerated sync output (updated timestamps across all tools). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * docs(architecture): add competitive landscape an…
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Brief description of what this PR does and why.
Closes #
Changes
Test Plan
Describe how you tested these changes:
Validation Commands
# Commands to verify this change works correctlyChecklist
npm test/cargo test/ equivalent)Documentation
Change Impact
Documentation Checklist (required for High impact)
./scripts/create-doc.sh <type> "<title>" <pr>[bracketed]placeholder sections filled indocs/history/subdirectory.index.jsonCHANGELOG.mdupdated (auto-updated bycreate-doc.sh, or run./scripts/update-changelog.sh)Summary by CodeRabbit
New Features
Bug Fixes
Documentation
Tests