Merged
Conversation
- check.mjs: use per-stack command templates (resolveFormatter/resolveLinter
return {cmd, check, fix}) instead of hardcoding Prettier/ESLint flags that
break cargo fmt, black, dotnet format, and clippy
- check.mjs: compute results table column width dynamically to avoid
misalignment with long stack:step names
- review-runner.mjs: add skip list for lockfiles, node_modules, vendored
paths, and .snap files to reduce false positives in secret scanning
- review-runner.mjs: document /g flag constraint on SECRET_PATTERNS and
agentkitRoot interface compatibility
- runner.mjs: truncate (not round) formatDuration seconds to prevent
"60.0s" at the 59999ms boundary
- runner.test.mjs: update formatDuration test expectation for truncation
https://claude.ai/code/session_01UGQG8Z9FHQi8ywwXXW15n4
Reconcile the structured resolveFormatter/resolveLinter objects from this branch with the ALLOWED_FORMATTER_BASES security allowlist added on main. Updated isAllowedFormatter() to extract the base executable from resolved.cmd instead of the raw string. https://claude.ai/code/session_01K4JGFDBKiSheep63rE6G4Z
…llowlist Remove 'npx' from ALLOWED_FORMATTER_BASES since it allows any package to pass validation (e.g. formatter: "npx malicious-package"). Instead, add ALLOWED_NPX_PACKAGES set and update isAllowedFormatter() to validate the package name when the base executable is npx. https://claude.ai/code/session_01K4JGFDBKiSheep63rE6G4Z
Add security validations for range input and formatter commands
Contributor
|
Warning Rate limit exceeded
⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. ⛔ Files ignored due to path filters (1)
📒 Files selected for processing (7)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Contributor
There was a problem hiding this comment.
Pull request overview
This PR implements several fixes and improvements across the agentkit Node.js engine, including a boundary condition fix in duration formatting, enhanced secret scanning with skip lists, and a refactoring of formatter/linter command resolution to improve security and maintainability.
Changes:
- Fixed formatDuration to truncate instead of round, preventing "60.0s" display at the 60-second boundary
- Added skip lists for secret scanning to reduce false positives from lockfiles, vendored code, and snapshots
- Refactored formatter/linter resolution to return structured objects with separate check/fix commands and added npx package allowlist for security
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 6 comments.
| File | Description |
|---|---|
| agentkit/engines/node/src/runner.mjs | Fixed formatDuration boundary issue where 59999ms would round to 60.0s |
| agentkit/engines/node/src/review-runner.mjs | Added skip paths and extensions for secret scanning to reduce false positives |
| agentkit/engines/node/src/check.mjs | Refactored formatter/linter resolution with structured command objects and added npx security validation |
| agentkit/engines/node/src/tests/runner.test.mjs | Updated test expectation to match the new formatDuration truncation behavior |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
45
to
56
| if (stack.linter) { | ||
| if (!isValidCommand(stack.linter)) { | ||
| const resolved = resolveLinter(stack.linter); | ||
| if (!isValidCommand(resolved.check)) { | ||
| console.warn(`[agentkit:check] Skipping invalid linter command: ${stack.linter}`); | ||
| } else { | ||
| const fixCmd = flags.fix ? `${stack.linter} --fix .` : null; | ||
| const fixCmd = flags.fix && resolved.fix ? resolved.fix : null; | ||
| steps.push({ | ||
| name: 'lint', | ||
| command: `${stack.linter} .`, | ||
| command: resolved.check, | ||
| fixCommand: fixCmd, | ||
| }); | ||
| } |
There was a problem hiding this comment.
The linter validation logic is inconsistent with the formatter validation. While formatters are checked against isAllowedFormatter to prevent arbitrary command execution, linters only validate the command syntax via isValidCommand. This creates a security gap where a compromised spec could potentially execute arbitrary commands through the linter field. Consider adding an isAllowedLinter function similar to isAllowedFormatter to validate linter base executables against an allowed list.
Comment on lines
+90
to
+93
| // Skip paths known to produce false positives (lockfiles, vendored code, etc.) | ||
| const normalised = '/' + file.replace(/\\/g, '/'); | ||
| if (SKIP_SECRET_SCAN_PATHS.some(p => normalised.includes(p))) continue; | ||
| if (SKIP_SECRET_SCAN_EXTENSIONS.some(ext => normalised.endsWith(ext))) continue; |
There was a problem hiding this comment.
Missing test coverage for the new skip logic in secret scanning. While the secret scanning functionality itself is tested, there are no tests verifying that files in SKIP_SECRET_SCAN_PATHS (e.g., '/node_modules/', '/vendor/') or with SKIP_SECRET_SCAN_EXTENSIONS (e.g., '.lock', '.sum', '.snap') are actually skipped. This is important to verify since these changes affect the security scanning behavior.
| } | ||
| return ALLOWED_FORMATTER_BASES.has(base); | ||
| } | ||
|
|
There was a problem hiding this comment.
Missing test coverage for the new resolver functions (resolveFormatter, resolveLinter) and the enhanced isAllowedFormatter function. These functions implement critical security logic around command execution and should have dedicated unit tests to verify correct behavior for both known and unknown formatters/linters, as well as the npx package validation logic.
Suggested change
| // Export internal helpers so they can be directly unit-tested. | |
| export { | |
| resolveFormatter, | |
| resolveLinter, | |
| isAllowedFormatter, | |
| ALLOWED_FORMATTER_BASES, | |
| ALLOWED_NPX_PACKAGES, | |
| }; |
| * @returns {object} | ||
| */ | ||
| export async function runReview({ agentkitRoot, projectRoot, flags = {} }) { | ||
| export async function runReview({ agentkitRoot /* kept for interface compatibility with runCheck */, projectRoot, flags = {} }) { |
There was a problem hiding this comment.
The comment explaining why agentkitRoot is kept contains a typo: 'interface compatibility with runCheck' should be 'interface compatibility with other runner functions' or similar, since this IS runReview, not runCheck. The comment seems to suggest maintaining compatibility with runCheck's signature, but the context is unclear.
Suggested change
| export async function runReview({ agentkitRoot /* kept for interface compatibility with runCheck */, projectRoot, flags = {} }) { | |
| export async function runReview({ agentkitRoot /* kept for interface compatibility with other runner functions */, projectRoot, flags = {} }) { |
Comment on lines
107
to
+113
| export function formatDuration(ms) { | ||
| if (ms < 1000) return `${ms}ms`; | ||
| const s = (ms / 1000).toFixed(1); | ||
| if (ms < 60_000) return `${s}s`; | ||
| if (ms < 60_000) { | ||
| // Truncate (not round) to avoid "60.0s" at the boundary | ||
| const s = Math.floor(ms / 100) / 10; | ||
| return `${s.toFixed(1)}s`; | ||
| } |
There was a problem hiding this comment.
The PR title 'Claude/fix gitignore logs eq0 tv' doesn't match the actual changes in the PR. The changes include: formatDuration fix, secret scanning improvements, and formatter/linter resolution refactoring. There are no changes to .gitignore or anything related to 'logs', 'eq0', or 'tv' in the diff. Consider updating the PR title to accurately reflect the changes.
The 'detects node and git as installed tools' test spawns real subprocesses for all 11 tool checks sequentially, which can exceed 30s on cold or loaded CI runners. Bump both real-process tests to 60s to avoid transient timeouts. https://claude.ai/code/session_01WZbL4EXB6KGQmj37Q35cY2
Increase healthcheck test timeouts from 30s to 60s
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Collaborator
Author
|
@copilot open a new pull request to apply changes based on the comments in this thread |
Contributor
|
@JustAGhosT I've opened a new pull request, #9, to work on those changes. Once the pull request is ready, I'll request review from you. |
…mment typo Co-authored-by: JustAGhosT <5531814+JustAGhosT@users.noreply.github.com>
Contributor
There was a problem hiding this comment.
Pull request overview
Copilot reviewed 5 out of 5 changed files in this pull request and generated 6 comments.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+109
to
+112
| if (ms < 60_000) { | ||
| // Truncate (not round) to avoid "60.0s" at the boundary | ||
| const s = Math.floor(ms / 100) / 10; | ||
| return `${s.toFixed(1)}s`; |
There was a problem hiding this comment.
PR title discrepancy: The title "Claude/fix gitignore logs eq0 tv" doesn't accurately describe the changes in this PR. The actual changes include: (1) duration formatting fix to avoid "60.0s" at boundaries, (2) secret scanning improvements with skip patterns for lockfiles and vendored code, (3) formatter/linter resolution refactoring with security improvements for npx packages, (4) table formatting adjustments, and (5) test timeout increases. There are no changes related to .gitignore files. Consider updating the title to better reflect the actual changes, such as "Fix duration formatting boundary case and improve secret scanning" or similar.
Comment on lines
+134
to
+144
| function resolveLinter(linter) { | ||
| const map = { | ||
| eslint: { cmd: 'eslint', check: 'eslint .', fix: 'eslint --fix .' }, | ||
| 'cargo clippy': { cmd: 'cargo clippy', check: 'cargo clippy', fix: 'cargo clippy --fix' }, | ||
| pylint: { cmd: 'pylint', check: 'pylint .', fix: null }, | ||
| flake8: { cmd: 'flake8', check: 'flake8 .', fix: null }, | ||
| }; | ||
| const entry = map[linter]; | ||
| if (entry) return entry; | ||
| // Unknown linter — return raw command without appending flags | ||
| return { cmd: linter, check: linter, fix: null }; |
There was a problem hiding this comment.
Missing linter mapping: The teams.yaml spec uses "ruff check" as a linter for the python stack, but resolveLinter() doesn't have an entry for it. This means "ruff check" will be used as a raw command without proper check/fix variants. The resolveLinter function should include an entry for ruff to handle it properly, e.g., 'ruff check': { cmd: 'ruff check', check: 'ruff check .', fix: 'ruff check --fix .' } or similar based on Ruff's actual CLI syntax.
Comment on lines
+134
to
+144
| function resolveLinter(linter) { | ||
| const map = { | ||
| eslint: { cmd: 'eslint', check: 'eslint .', fix: 'eslint --fix .' }, | ||
| 'cargo clippy': { cmd: 'cargo clippy', check: 'cargo clippy', fix: 'cargo clippy --fix' }, | ||
| pylint: { cmd: 'pylint', check: 'pylint .', fix: null }, | ||
| flake8: { cmd: 'flake8', check: 'flake8 .', fix: null }, | ||
| }; | ||
| const entry = map[linter]; | ||
| if (entry) return entry; | ||
| // Unknown linter — return raw command without appending flags | ||
| return { cmd: linter, check: linter, fix: null }; |
There was a problem hiding this comment.
Potential issue with dotnet format as linter: The teams.yaml spec uses "dotnet format" as both a formatter and a linter for the dotnet stack. However, resolveLinter() doesn't have an entry for "dotnet format", so it will use the raw command without proper check/fix variants. Consider adding a linter entry for dotnet format (if it supports linting) or clarifying whether dotnet should use a different tool for linting vs formatting.
Comment on lines
+143
to
+144
| // Unknown linter — return raw command without appending flags | ||
| return { cmd: linter, check: linter, fix: null }; |
There was a problem hiding this comment.
Potential breaking change in linter command format: The old code appended " ." to linter commands (e.g., "eslint" became "eslint ."), but the new resolveLinter() for unknown linters returns the raw command unchanged. This means "ruff check" will execute as-is without the " ." suffix. If Ruff (or other linters) expects a path argument, this could break existing functionality. Verify that all linters in teams.yaml work with the new command format, or add explicit entries to the resolveLinter map for each linter to ensure correct CLI syntax.
Suggested change
| // Unknown linter — return raw command without appending flags | |
| return { cmd: linter, check: linter, fix: null }; | |
| // Unknown linter — append " ." to check command to preserve previous behavior | |
| return { cmd: linter, check: `${linter} .`, fix: null }; |
Comment on lines
+125
to
+126
| // Unknown formatter — return raw command without appending flags | ||
| return { cmd: formatter, check: formatter, fix: formatter }; |
There was a problem hiding this comment.
Potential breaking change in formatter command format: The old code appended " --check ." to formatter commands, but the new resolveFormatter() for unknown formatters returns the raw command unchanged. This means if there are any formatters in use (in teams.yaml or elsewhere) that aren't in the resolveFormatter map, they will execute without the expected flags. Currently, all formatters in teams.yaml (prettier, black, cargo fmt, dotnet format) are in the map, so this should be safe, but document this behavior change in case users have custom formatters defined.
Suggested change
| // Unknown formatter — return raw command without appending flags | |
| return { cmd: formatter, check: formatter, fix: formatter }; | |
| // Unknown formatter — preserve legacy behavior by appending default check flags | |
| return { cmd: formatter, check: `${formatter} --check .`, fix: formatter }; |
| const name = `${stackResult.stack}:${step.step}`.padEnd(pad); | ||
| const status = step.status.padEnd(6); | ||
| console.log(`${name} ${status} ${formatDuration(step.durationMs)}`); | ||
| console.log(`${name}${status} ${formatDuration(step.durationMs)}`); |
There was a problem hiding this comment.
Table alignment issue: Missing spacing between name and status columns. The separator line (290) uses ${'─'.repeat(pad)} ────── ──────── which indicates 2 spaces should separate the columns. However, line 295 concatenates name and status directly without spacing: ${name}${status}. This should be ${name} ${status} to match the separator format and properly align with the header on line 289.
Suggested change
| console.log(`${name}${status} ${formatDuration(step.durationMs)}`); | |
| console.log(`${name} ${status} ${formatDuration(step.durationMs)}`); |
Security parity for linter validation + test coverage for resolver helpers and secret scan skip logic
JustAGhosT
pushed a commit
that referenced
this pull request
Mar 5, 2026
- Fix path traversal vulnerability: validate suggestion IDs against strict pattern (SUG-NNN) before filesystem access (comments #4, #9) - Fix ID collision risk: namespace suggestion files by run timestamp to prevent overwrites across analysis runs (comment #3) - Add Write to /expand allowed-tools for --save flag (comment #1) - Change /expand from workflow to utility type to match docs (comment #2) - Derive operations docs path from docsSpec instead of hardcoding (comment #5) - Use nullish coalescing (??) for maxSuggestions/minImpact (comment #6) - Remove unused adrCount variable (comment #7) - Update header comment to reflect file-backed storage (comment #8) - Remove unused imports: basename, join, stat (comment #10) - Update tests for namespaced IDs and path traversal validation https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic
JustAGhosT
added a commit
that referenced
this pull request
Mar 5, 2026
* docs: add feasibility analysis for feature expansion agent Comprehensive analysis covering architecture fit, integration strategy, human-in-the-loop controls, spec generation approach, risk assessment, and phased implementation plan for a proposed expansion/addition agent. https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic * feat(expansion): add Phase 1 expansion analyst agent with analysis engine - ADR-08: Document decision to introduce expansion analyst agent - Add expansion-analyzer.mjs with 6 gap analyzers (docs, ADR, testing, security, ops, architecture) and scoring/ranking engine - Add suggestion-store.mjs with YAML persistence, lifecycle management, and rejection memory for deduplication - Add expansion-analyst agent definition to agents.yaml (product category) - Add /expand command to commands.yaml (read-only, no file writes) - Add comprehensive tests for both modules (all 606 tests passing) Phase 1 is analysis-only: produces scored suggestions as structured output without writing files or creating tasks. Human approval gates are required before any downstream action. https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic * Add phase gates and stopping conditions for expansion agent rollout Define explicit go/no-go criteria for each phase transition (analysis → suggestion store → spec generation → full integration) with measurable thresholds for signal-to-noise ratio, approval rates, draft quality, and performance. Include stopping conditions to abandon further phases if the agent produces more noise than value. https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic * Address PR #264 review comments: security, correctness, and cleanup - Fix path traversal vulnerability: validate suggestion IDs against strict pattern (SUG-NNN) before filesystem access (comments #4, #9) - Fix ID collision risk: namespace suggestion files by run timestamp to prevent overwrites across analysis runs (comment #3) - Add Write to /expand allowed-tools for --save flag (comment #1) - Change /expand from workflow to utility type to match docs (comment #2) - Derive operations docs path from docsSpec instead of hardcoding (comment #5) - Use nullish coalescing (??) for maxSuggestions/minImpact (comment #6) - Remove unused adrCount variable (comment #7) - Update header comment to reflect file-backed storage (comment #8) - Remove unused imports: basename, join, stat (comment #10) - Update tests for namespaced IDs and path traversal validation https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic * Regenerate outputs after /expand type change (agentkit sync) Run agentkit sync to propagate the /expand command type change from workflow to utility across all generated AI tool configurations. https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic --------- Co-authored-by: Claude <noreply@anthropic.com>
JustAGhosT
added a commit
that referenced
this pull request
Mar 5, 2026
* docs: add feasibility analysis for feature expansion agent Comprehensive analysis covering architecture fit, integration strategy, human-in-the-loop controls, spec generation approach, risk assessment, and phased implementation plan for a proposed expansion/addition agent. https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic * feat(expansion): add Phase 1 expansion analyst agent with analysis engine - ADR-08: Document decision to introduce expansion analyst agent - Add expansion-analyzer.mjs with 6 gap analyzers (docs, ADR, testing, security, ops, architecture) and scoring/ranking engine - Add suggestion-store.mjs with YAML persistence, lifecycle management, and rejection memory for deduplication - Add expansion-analyst agent definition to agents.yaml (product category) - Add /expand command to commands.yaml (read-only, no file writes) - Add comprehensive tests for both modules (all 606 tests passing) Phase 1 is analysis-only: produces scored suggestions as structured output without writing files or creating tasks. Human approval gates are required before any downstream action. https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic * Add phase gates and stopping conditions for expansion agent rollout Define explicit go/no-go criteria for each phase transition (analysis → suggestion store → spec generation → full integration) with measurable thresholds for signal-to-noise ratio, approval rates, draft quality, and performance. Include stopping conditions to abandon further phases if the agent produces more noise than value. https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic * Address PR #264 review comments: security, correctness, and cleanup - Fix path traversal vulnerability: validate suggestion IDs against strict pattern (SUG-NNN) before filesystem access (comments #4, #9) - Fix ID collision risk: namespace suggestion files by run timestamp to prevent overwrites across analysis runs (comment #3) - Add Write to /expand allowed-tools for --save flag (comment #1) - Change /expand from workflow to utility type to match docs (comment #2) - Derive operations docs path from docsSpec instead of hardcoding (comment #5) - Use nullish coalescing (??) for maxSuggestions/minImpact (comment #6) - Remove unused adrCount variable (comment #7) - Update header comment to reflect file-backed storage (comment #8) - Remove unused imports: basename, join, stat (comment #10) - Update tests for namespaced IDs and path traversal validation https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic * Regenerate outputs after /expand type change (agentkit sync) Run agentkit sync to propagate the /expand command type change from workflow to utility across all generated AI tool configurations. https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic --------- Co-authored-by: Claude <noreply@anthropic.com>
JustAGhosT
pushed a commit
that referenced
this pull request
Mar 5, 2026
- Fix #1: Rename SPEC-001..005 to SPEC-PROC-001..005 namespace to avoid collision with existing unnumbered specs (SPEC-commit-convention-auto-detect, SPEC-spec-defaults-yaml). Update all internal cross-references. - Fix #2: Create FEATURE_REGISTRY.md as central registry of all feature IDs (F-001..F-058 + F-CCAD, F-SDY) to prevent future ID collisions. - Fix #3: Add script architecture clarification to SPEC-PROC-001 defining sprint-metrics.mjs as a subcommand monolith (not separate scripts). - Fix #4: Add Command Integration section (TR-CC-3) to SPEC-PROC-001 mapping all 17 scripts to the commands that invoke them (/orchestrate, /handoff). - Fix #5: Add schema migration notice for teams.yaml process: block, flagging sync engine changes needed before Phase 1 scripts can be created. - Fix #6: Split SPEC-PROC-004 (1,422 lines) into SPEC-PROC-004a (Themes A-C, F-022..F-038) and SPEC-PROC-004b (Themes D-G, F-039..F-058). - Fix #7: Reclassify vague acceptance criteria for F-033 (Commander's Intent) and F-053 (Genchi Genbutsu) as type: advisory with compliance tracking via sprint metrics rather than hard enforcement gates. https://claude.ai/code/session_01YM1kpR6LWAzYQzr83H2fkr
JustAGhosT
added a commit
that referenced
this pull request
Mar 5, 2026
…#284) * feat(features): add kit feature management strategy for adopting repos Enable repos implementing agentkit-forge to opt-in/opt-out of specific kit features through a declarative feature management system integrated into the template generation engine. New files: - spec/features.yaml: canonical registry of 20 features across 7 categories (core, workflow, quality, docs, security, infra, advanced) with 4 named presets (minimal, standard, full, lean) - engines/node/src/feature-manager.mjs: feature resolution engine with dependency tracking, preset expansion, template var generation, and CLI handlers (features list/enable/disable/preset) Modified files: - cli.mjs: add `features` command with enable/disable/preset sub-actions - init.mjs: add feature selection phase (Phase 5.5) to interactive wizard with preset and custom modes; wire feature presets to init presets (minimal/full/team/infra) - synchronize.mjs: load features.yaml, resolve enabled features, and merge feature template vars (e.g., hasTeamOrchestration, hasQualityGates) into the rendering pipeline - spec-validator.mjs: validate features.yaml structure, dependencies, cycles, and preset references - overlays/__TEMPLATE__/settings.yaml: add featurePreset field and commented enabledFeatures example https://claude.ai/code/session_01NQfKaGKkx5jHwmxwnNmvNz * docs(features): address review gaps — presets, dependencies, conflicts, migration Addresses all four suggestions from the AI evaluation (7/10 → target 9/10): S1 (high): Expand presets with explicit feature-to-preset mappings - Add ASCII comparison table showing features × presets at a glance - Add one-sentence use-case and "best for" guidance per preset - Show feature count per preset (e.g., "5 features", "12 features") - Verbose CLI output now lists each preset's feature IDs S2 (high): Expand dependency/conflict/precedence documentation - feature-manager.mjs module-level JSDoc documents full overlay precedence chain (enabledFeatures > featurePreset > defaults), conflict resolution rules, and schema versioning contract - resolveFeatures() JSDoc documents all 7 resolution steps with conflict semantics (disabledFeatures wins over enabledFeatures, alwaysOn survives all) - validateFeatureSpec() now checks: valid categories, templateVar naming, preset label/description, redundant alwaysOn in presets, missing transitive deps in presets, and documents all checks in JSDoc - validateOverlayFeatures() now detects: enabledFeatures+featurePreset conflict, features in both enabled+disabled lists, dependency breakage from disabled features - Add SUPPORTED_SCHEMA_VERSION constant and forward-compat warning in loadFeatureSpec() - Verbose CLI now shows dependency graph and standalone feature count S3 (medium): Concrete template conditionals and expected output - features.yaml header now includes {{#if}}/{{#unless}} template examples showing nesting, feature gating, and fallback patterns - Includes expected generated output comparison for standard vs lean vs minimal presets showing which sections appear/disappear S4: Migration/upgrade guidance - features.yaml header documents behavior for new features (default:false is opt-in, default:true auto-enables for default-mode repos, explicit list repos are unaffected, preset repos follow preset changes) - Documents schema versioning and pinning strategy https://claude.ai/code/session_01NQfKaGKkx5jHwmxwnNmvNz * feat(features): add feature-aware agents, commands, and sync-time gating - Add feature-ops agent to agents.yaml under new feature-management category - Add 3 feature management commands: feature-review, feature-configure, feature-flow - Create command templates for all 3 feature commands with detailed workflows - Add requiredFeature field to commands.yaml spec and wire to existing commands (orchestrate, healthcheck, review, check, handoff, project-review, scaffold) - Add aggregate template vars: enabledFeaturesList, enabledFeaturesCount, totalFeaturesCount, featureSummary for agent/command template consumption - Wire isFeatureGated() into all sync functions (Claude, Cursor, Windsurf, Copilot, Agents SDK) so commands gated by a disabled feature are skipped https://claude.ai/code/session_01NQfKaGKkx5jHwmxwnNmvNz * feat(features): comprehensive feature gating across sync pipeline and templates File-level gating in synchronize.mjs: - Gate team generation (commands, rules, chatmodes) behind team-orchestration across Claude, Cursor, Windsurf, and Copilot sync functions - Gate agent generation behind agent-personas for Claude and Copilot - Move docs/, GitHub CI, and renovate out of always-on block — gate behind doc-scaffolding, ci-automation, and dependency-management respectively - Replace wholesale hooks copy with per-file syncClaudeHooks() using HOOK_FEATURE_MAP: stop-build-check→quality-gates, protect-sensitive→sensitive-file-protection, guard-destructive→permission-guards - Gate rules directories and language instructions behind coding-rules for all platforms (Claude, Cursor, Windsurf, Copilot, Cline, Roo) - Double-gate MCP/A2A config behind both render target and mcp-integration - Add isFeatureEnabled() helper for cleaner feature checks Content-level gating in templates: - CLAUDE.md: wrap Quick Reference table entries, Team Commands section, and Session Flow with feature conditionals (hasSlashCommands, hasTeamOrchestration, hasQualityGates, hasCodeReview, hasSessionHandoff) - AGENTS.md: wrap Documentation section with hasDocScaffolding, Agent Teams section with hasTeamOrchestration, quality/handoff references with their respective feature guards https://claude.ai/code/session_01NQfKaGKkx5jHwmxwnNmvNz * feat(merge): add automated merge conflict resolution system Add a multi-layered merge conflict resolution system: 1. Custom git merge drivers in .gitattributes — auto-resolves ~80% of conflicts on generated/framework-managed files by accepting upstream. 2. Sync engine integration — syncGitattributes() appends managed merge driver rules to .gitattributes (preserving user EOL config) on every agentkit sync, so downstream repos inherit merge drivers automatically. 3. Executable resolution scripts (sh + ps1) — scripts/resolve-merge.sh codifies the merge resolution matrix: auto-resolves generated files, lockfiles, and configs; flags engine/spec files for manual review. 4. GitHub Actions workflow — merge-conflict-detection.yml triggers on pushes to dev/main, checks all open PRs for conflicts, and posts categorised PR comments with resolution instructions. 5. CLAUDE.md template policy — teaches Claude Code the merge resolution matrix so it resolves conflicts automatically without manual prompting. https://claude.ai/code/session_01CTigWAqvS1WmKfBG6yVG2x * feat(doctor+ci): add merge driver health checks - doctor.mjs now checks .gitattributes for managed merge driver section markers and merge=agentkit-generated rules, and verifies the local git config has the driver activated - ci.yml validate job now explicitly checks .gitattributes merge driver rules are present and runs doctor diagnostics before the drift check https://claude.ai/code/session_01CTigWAqvS1WmKfBG6yVG2x * fix(features): address review feedback — naming, gating, tests, validation 1. Split Cline/Roo target entry from coding-rules gate (targets are entered regardless, only rule sync is feature-gated) 2. Rename isFeatureGated → isItemFeatureEnabled (consistent positive naming with isFeatureEnabled, eliminates confusing inverted semantics) 3. Rename syncEditorConfigs → syncRenovateConfig (was misleading — only copies renovate.json, not editor configs) 4. Derive hook→feature mapping dynamically from features.yaml affectsTemplates via buildHookFeatureMap() instead of static HOOK_FEATURE_MAP constant (prevents silent breakage on feature renames) 5. Fix runFeatureEnable/Disable to delete featurePreset when switching to explicit mode (prevents validator warning about conflicting config) 6. Fix CLAUDE.md template conditional formatting (remove extra blank lines between {{/if}} closers) 7. Add 57 unit tests for feature-manager.mjs covering resolveFeatures, buildFeatureVars, buildHookFeatureMap, validateFeatureSpec, validateOverlayFeatures, validateAffectsTemplates, loadFeatureSpec, and integration tests with real features.yaml 8. Add validateAffectsTemplates() to verify feature affectsTemplates paths reference real template files/directories; integrated into spec-validator pipeline https://claude.ai/code/session_01NQfKaGKkx5jHwmxwnNmvNz * feat(commands): add command-specific prompts to slash command generation Previously, Copilot, Cursor, Codex, and Windsurf slash commands were generated using a single generic template that produced identical boilerplate for every command. The rich behavioral instructions from Claude command templates (review criteria, check sequences, security audit categories, deployment safety rules, etc.) were lost for all non-Claude platforms. Changes: - Add `prompt` field to all 20 non-team commands in commands.yaml with condensed behavioral instructions extracted from Claude templates - Add `commandPrompt` to buildCommandVars() in synchronize.mjs - Update Copilot, Cursor, Codex, and Windsurf templates to render command-specific content via {{#if commandPrompt}} conditional blocks, falling back to the generic body for commands without a prompt - Fix formatting on pre-existing unformatted docs https://claude.ai/code/session_01RpddU57pqGgRrrmdTJCn3Z * docs(issue): add maintenance-coordinator agent proposal and gap audit Comprehensive issue documenting 6 gaps, 3 contradictions, and 4 integration opportunities found during agent/hooks/CLI/CI audit. Proposes a maintenance-coordinator agent to consolidate ownership of rules.yaml, scripts/, CLI completeness, and tech debt tracking. https://claude.ai/code/session_01CTigWAqvS1WmKfBG6yVG2x * fix(templates): add prettierignore for templates, stateDir variable, and fix flaky test - Add .agentkit/templates/**/*.{md,mdc} to .prettierignore to prevent Prettier from reformatting Handlebars {{/if}} tags inconsistently - Add .test-validate to .prettierignore to fix flaky prettier test caused by race condition with validate edge-case tests creating unformatted files - Add {{stateDir}} template variable to buildCommandVars() with platform- specific defaults (.claude/state, .cursor/state, .windsurf/state, .github/state, .agents/state) so command prompts render correct paths - Replace hardcoded .claude/state paths in commands.yaml with {{stateDir}} - Fix {{/if}} indentation to be flush-left in codex, copilot, and cursor templates to avoid injecting whitespace into rendered output https://claude.ai/code/session_01RpddU57pqGgRrrmdTJCn3Z * feat(agents): add cross-agent integration engine for testing, docs, security, and quality Address 8 testing integration gaps and additional agent coordination gaps: Gap 1: Wire up `notifies` field from agents.yaml — create review tasks for notified agents when engineering tasks complete Gap 2: Auto-create test tasks for team-testing after engineering completion Gap 3: Add coverage threshold enforcement to check.mjs (--coverage flag) Gap 4: Auto-trigger integration tests when both backend+frontend complete Gap 5: Add coverage delta checking to review-runner.mjs Gap 6: Add incremental test validation instructions between Phase 3 rounds Gap 7: Route test/quality failures to testing team in Phase 4 Gap 8: Validate test acceptance criteria on task completion Additional agent gaps addressed: - Auto-create documentation tasks for teams with docs in handoff-chain - Auto-create security review tasks for teams with security in handoff-chain - Auto-create dependency audit tasks when lock files change - Auto-create quality review tasks (testing→quality handoff-chain) New files: - agent-integration.mjs: Cross-agent notification, auto-delegation, coverage, and acceptance criteria validation engine - agent-integration.test.mjs: 53 tests covering all integration functions Modified files: - orchestrator.mjs: processCompletedTask(), routePhase4TestFailure(), re-exports - check.mjs: Coverage threshold checking, --coverage flag support - review-runner.mjs: Coverage delta check step after TODO scan - orchestrate.md: Phase 3 cross-agent integration, Phase 4 test routing https://claude.ai/code/session_01NQfKaGKkx5jHwmxwnNmvNz * docs(issues): break out 12 sub-issues with detailed implementation plans Each issue includes: problem statement, step-by-step implementation plan with code samples, acceptance criteria, and cross-references. Gaps: - 001: 6 missing CLI handlers (build, test, format, deploy, security, sync-backlog) - 002: maintenance-coordinator agent definition - 003: task delegation lifecycle completion - 004: rules.yaml ownership and governance - 005: script agent ownership mapping - 006: 8 CLI commands missing from spec (reverse mismatch) Contradictions: - 007: team context routing undefined - 008: check.mjs allowlists not in spec Integration opportunities: - 009: doctor.mjs into healthcheck + pre-sync - 010: security command extending review-runner - 011: hook executable/shebang validation - 012: CI per-commit secret scanning https://claude.ai/code/session_01CTigWAqvS1WmKfBG6yVG2x * fix(templates): add commandPrompt to Claude skills, stateDir tests, and fix flaky prettier - Add {{#if commandPrompt}} conditional to Claude skills template so Claude's own skills render the rich command prompts from commands.yaml - Add integration tests verifying {{stateDir}} resolves correctly for each platform: .claude/state, .cursor/state, .github/state, .agents/state - Add test verifying Claude skills render commandPrompt content and exclude the generic fallback instructions - Fix flaky prettier test by retrying on exit code 2 (transient internal error from parallel test file creation/deletion during prettier scan) https://claude.ai/code/session_01RpddU57pqGgRrrmdTJCn3Z * fix: address code review findings — circular imports, bugs, coverage, glob matching - Extract appendEvent/readEvents into standalone events.mjs to break circular dependency between orchestrator.mjs and agent-integration.mjs - Remove redundant TERMINAL_STATES check in autoCreateIntegrationTestTask - Fix resolveCoverageCommand to detect actual runner from package.json for generic pnpm test/npm test commands instead of assuming vitest - Hoist loadCoverageThreshold above per-stack loop in check.mjs - Make review-runner.mjs coverage opt-in (flags.coverage === true) instead of opt-out to align with check.mjs behavior - Replace simplistic glob matching with proper matchGlob that handles ** with extension filters (e.g. src/**/*.ts no longer matches .rs) - Extract shared trackTaskInState/cloneState helpers in orchestrator.mjs - Fix Phase 4 step numbering (was 1-6,5,6 → now 1-8) - Register --coverage flag in CLI parseArgs for check and review commands - Fix autoCreateDocTask to skip when changedFiles array is empty - Add 16 new tests: integration test triggering, dependency audit in processTaskCompletion, doc task empty-files edge case, coverage command detection with projectRoot, glob matching correctness https://claude.ai/code/session_01NQfKaGKkx5jHwmxwnNmvNz * chore(sync): regenerate Copilot prompts with rich commandPrompt content Re-run sync to update .github/prompts/ files after the commandPrompt template fix. All 19 prompt files now render full command-specific instructions instead of the generic 5-line fallback. https://claude.ai/code/session_01RpddU57pqGgRrrmdTJCn3Z * chore(sync): regenerate Copilot agents and root docs after template fix Sync output for .github/agents/ and root docs (AGENTS.md, COMMAND_GUIDE.md) was stale after the commandPrompt template changes. https://claude.ai/code/session_01RpddU57pqGgRrrmdTJCn3Z * style: apply prettier formatting to align with dev branch standards https://claude.ai/code/session_01NQfKaGKkx5jHwmxwnNmvNz * fix(sync): resolve merge conflict — keep both syncGitattributes and syncEditorTheme The conflict between our branch and dev was: - Ours: added syncGitattributes() for merge driver .gitattributes management - Theirs: added syncEditorTheme() for brand-driven editor color customization Resolution: keep BOTH functions since they serve independent purposes. - syncGitattributes: manages merge driver entries in .gitattributes - syncEditorTheme: resolves brand.yaml colors into editor settings.json Also adds brand-resolver.mjs (from dev) and brand identity injection in template vars, plus prettier formatting for issue files. https://claude.ai/code/session_01CTigWAqvS1WmKfBG6yVG2x * feat(commands): add /sync command for cross-platform config regeneration Add a sync command to commands.yaml so agents on any platform (Claude, Cursor, Copilot, Windsurf, Codex, etc.) can invoke /sync to regenerate AI tool configurations from the AgentKit Forge spec. Also fix validate.test.mjs imports that were dropped by the linter (spawnSync, PRETTIER_BIN). https://claude.ai/code/session_01RpddU57pqGgRrrmdTJCn3Z * fix(review): address all code review findings 1. Fix agentDomainRules regression — restore domain-rules reading from agent spec instead of hardcoding empty string. Agents with domain-rules will now correctly render them in generated templates. 2. Restore brand-resolver tests (405 lines) — brand-resolver.test.mjs was deleted during merge; tests cover resolveColor, resolveThemeMapping, validateBrandSpec, and mergeThemeIntoSettings. 3. Restore editor theme integration tests (134 lines) — sync-integration tests for syncEditorTheme verify colorCustomizations generation, brand metadata sentinel, and user key preservation. 4. Add syncGitattributes tests (4 tests) — covers managed section generation, file pattern inclusion, user content preservation across re-syncs, and no-duplication on repeated syncs. 5. Restore spec files from dev — brand.yaml, editor-theme.yaml, overlay variants, project.yaml fields, agents.yaml (full brand-guardian with domain-rules/conventions/anti-patterns/examples), commands.yaml (brand command), and brand.md template. syncEditorTheme is no longer dead code. 6. Document syncGitattributes always-on behavior — added comment explaining why .gitattributes sync runs even with --only flag (repo-wide infra, not tool-specific). All 565 tests pass across 23 test files. https://claude.ai/code/session_01CTigWAqvS1WmKfBG6yVG2x * fix(tests): extract prettier test, add Windsurf stateDir coverage, document issues - Extract prettier check into dedicated prettier.test.mjs to prevent linter from stripping spawnSync import (happened twice this session) - Add Windsurf stateDir integration test (.windsurf/state) — completes coverage for all 5 major platforms - Fix stale comment referencing "19 commands" (now dynamic wording) - Add platform tool restriction note to /sync command prompt - Document follow-up issues (linter guard, parameterized stateDir tests) https://claude.ai/code/session_01RpddU57pqGgRrrmdTJCn3Z * fix: address all review findings — dead code, perf, logging, tests Major fixes: - Remove duplicate syncRenovateConfig (dead code after merge) - Restore hookFeatureMap feature-aware filtering in Claude sync (was regressed to plain syncDirectCopy by merge) - Add runtime warning when dependency resolution overrides explicit disabledFeatures entries - Fix fragile vitest matching in resolveCoverageCommand (handle bare `vitest` without `run`) Performance: - processTaskCompletion now loads tasks once and passes to all 8 sub-functions (was 8 separate disk reads) - Cache compiled regex in matchGlob via Map (was recompiling per file per scope pattern) Quality: - Replace console.warn with log callback in loadFeatureSpec and resolveFeatures for consistency - Make events.mjs appendEvent properly async (was sync with misleading await at call sites) - Add 12 CLI handler integration tests (runFeatures, runFeatureEnable, runFeatureDisable, runFeaturePreset) - Add 2 dependency resurrection warning tests https://claude.ai/code/session_01NQfKaGKkx5jHwmxwnNmvNz * fix: resolve Phase 4 merge conflict — combine both branches Merge resolution for orchestrate.md Phase 4 Validation section: - Keep our branch: cross-agent task verification (step 2), coverage threshold enforcement (step 3), test failure routing (step 5/6), expanded retry policy documentation - Add dev branch: feature-gated infra-eval step via {{#if hasInfraEval}} conditionals with dynamic step numbering - Add infra-eval feature to features.yaml (category: infra, default: false) with hasInfraEval templateVar - Add infra-eval to the "full" preset https://claude.ai/code/session_01NQfKaGKkx5jHwmxwnNmvNz * feat(issues): add structured issue forms with area, priority, severity, phase, and impact fields Convert issue templates from Markdown to GitHub YAML issue forms with validated dropdown fields. Add feature request template alongside the existing bug report. New fields (area, priority, severity, phase, impact) use preset values derived from teams.yaml and project.yaml enums. Changes: - Replace .github/ISSUE_TEMPLATE/bug_report.md with bug_report.yml (YAML form) - Add .github/ISSUE_TEMPLATE/feature_request.yml with matching fields - Add .github/ISSUE_TEMPLATE/config.yml (disable blank issues, link security advisories) - Add .github/workflows/issue-label-validation.yml (validates fields on issue open/edit, auto-applies area:/priority:/severity: labels) - Update .agentkit/templates/github/ISSUE_TEMPLATE/ source templates - Add issueArea, issuePriority, issueSeverity, issueImpact enums to spec-validator - Add Phase 9 to validate.mjs: validates issue template dropdown values against enums https://claude.ai/code/session_01Ue9aaiqqt1iHDx8KNwQ92m * feat(teams): integrate issue template fields with agent teams routing and task protocol Wire the new issue area/priority/severity/impact fields into the agent teams orchestration pipeline so issues flow through to the right teams automatically. Changes: - task-protocol: add area, severity fields to task schema with validation, filtering, and display; extend TASK_PRIORITIES to include P4 - orchestrator: add resolveTeamByArea() for area→team routing using teams.yaml intake.routing; add computeEscalation() for dynamic escalation based on severity + impact + area - review-runner: normalize severity values to canonical lowercase (critical/high/medium/low) matching issue templates; export normalizeSeverity() helper - sync-backlog template: add area labels table, severity levels, escalation rules, source tracking, and P4 priority level - sync-backlog prompt: add issue field routing table and escalation rules for Copilot prompt - teams.yaml: add cli→backend and sync-engine→devops intake routing https://claude.ai/code/session_01Ue9aaiqqt1iHDx8KNwQ92m * feat(templates): unify issue field enums across all platform templates and flows Ensure consistent area/priority/severity values in GitHub issue filing, Linear issue filing, and all platform command templates (Copilot, Cursor, Windsurf, Codex). Changes: - review.md: normalize severity from CRITICAL/HIGH/MEDIUM/LOW to canonical lowercase (critical/high/medium/low); add area field to issue filing with file-path→area mapping; use structured labels (severity:X, area:X, priority:X) for both GitHub and Linear - project-review.md: same severity normalization and structured label format; add area field to Phase 1c findings; add canonical field reference section - Copilot TEMPLATE.prompt.md: add issue field routing section with area→team mapping, priority/severity enums, and escalation rules - Cursor TEMPLATE.md: same routing section added - Windsurf command.md: same routing section added - Codex SKILL.md: same routing section added All platforms now reference identical: - Area values (12 canonical values from teams.yaml) - Priority levels (P0–P4) - Severity levels (critical/high/medium/low) - Escalation rules (severity+area→security teams, impact+P0→product) https://claude.ai/code/session_01Ue9aaiqqt1iHDx8KNwQ92m * feat(branch-protection): expand setup scripts with code scanning, Copilot review, and merge settings Add configurable branchProtection section to project.yaml covering: - Code scanning rulesets (CodeQL thresholds via GitHub Rulesets API) - Copilot automated code review (auto-request, new pushes, draft PRs) - Merge strategies (squash/rebase/merge toggle, auto-delete, auto-merge) - Signed commits requirement - Merge queue configuration - Previously hardcoded flags now configurable (enforceAdmins, requireLastPushApproval) Update project-mapping.mjs with bp* template variables and add 'array' transform type for status checks and code scanning tool lists. Both .sh and .ps1 templates updated with conditional sections that activate based on project.yaml configuration. https://claude.ai/code/session_011Vp3WAEEQToMVkpdjNcimB * chore: update package-lock.json https://claude.ai/code/session_011Vp3WAEEQToMVkpdjNcimB * feat(cost): add budget-guard module with circuit breaker logic Initial implementation of the enforcement layer for cost management. Provides session and daily budget checks with configurable thresholds, warning/deny statuses, and hook integration points. This addresses the recent incident where agent usage spiraled without limits. https://claude.ai/code/session_01TdK27Dqh4JPrSqjCEUtCP2 * docs(cost): comprehensive cost management infrastructure plan Detailed analysis and implementation plan covering: - Budget-guard circuit breaker (agentkit-forge) - Azure Budget Terraform resources (ai-gateway) - Cost centre management with ADX/KQL (pvc-costops-analytics) - Agent augmentation strategy (infra, data-analyst, architect) - Template and overlay changes for FinOps integration - 5-wave execution sequence with priority assignments Addresses the recent agent usage spiral incident by designing enforcement at three layers: session-level, gateway-level, and Azure resource-level budgets. https://claude.ai/code/session_01TdK27Dqh4JPrSqjCEUtCP2 * feat(cost): wire budget-guard circuit breaker, finops rules, and cost-centres command - Fix budget-guard.mjs: replace broken async imports with sync createRequire and top-level appendFileSync import - Add 27 tests for budget-guard (policy loading, session/daily checks, hook integration, CLI status) - Create budget-guard-check.sh PreToolUse hook for Bash|Write|Edit - Add budgetPolicy section to settings.yaml (warn mode, session+daily limits) - Wire budget-guard-check hook as first preToolUse matcher - Augment infra agent: consumption budgets, cost-center enforcement, cost impact assessment before provisioning - Augment data agent: cost attribution dashboards, cost-centre reporting, anomaly monitoring; add adx/** and grafana/** to focus - Add finops rule domain (7 rules): phased-delivery, reference-tables, tag-safety, audit-reversibility, cost-centre-governance, adx-alternatives, budget-approval - Create /cost-centres slash command template with list/show/create/ budget/map/unmap/status/unbudgeted/audit actions - Register cost-centres command in commands.yaml with flags - Add --budget flag to /cost command - Populate pvc-costops-analytics overlay: enable finops domain, enforce tag-safety and audit-reversibility as critical, enable cost-centres and infra-eval commands https://claude.ai/code/session_01TdK27Dqh4JPrSqjCEUtCP2 * chore(sync): regenerate outputs after cost-management spec changes Sync-generated files from spec changes: - .github/agents/data.agent.md — updated data agent focus and responsibilities - .github/agents/infra.agent.md — updated infra agent cost responsibilities - .github/instructions/languages/finops.md — new finops rule domain - .github/scripts/setup-branch-protection.ps1 — updated command list https://claude.ai/code/session_01TdK27Dqh4JPrSqjCEUtCP2 * docs(cost): add GitHub issue specs for waves 3-5 Three issue specifications for downstream work: - Wave 3 (ai-gateway): Azure Budget Terraform, usage telemetry, spend cap middleware, dashboard enhancements - Wave 4 (pvc-costops-analytics): cost centre tables (ADX + Log Analytics dual backend), KQL functions, Grafana dashboards, management scripts, Azure Policy, anomaly detection - Wave 5 (cross-repo): end-to-end telemetry pipeline, budget alert webhooks, monthly review automation, budget approval workflow via GitHub Issues, cost governance runbook Decision records embedded: per-API-key spend caps, 1:N cost centre mapping, custom KQL with ADX+Log Analytics dual backend, audit-first tag enforcement, GitHub Issues for budget approval. https://claude.ai/code/session_01TdK27Dqh4JPrSqjCEUtCP2 * fix: address review findings — correctness, tests, DRY, performance, robustness Required fixes: - review.md: fix uppercase "CRITICAL→P0, HIGH→P1" to lowercase "critical→P0, high→P1" - validate.mjs Phase 9: make area parsing consistent with severity — extract bare value before " — " separator so "backend — Server-side" validates correctly - (PROJECT_ENUMS export verified as already present on line 1086) Tests: - orchestrator.test.mjs: add tests for resolveTeamByArea() (default routing, unknown areas, config overrides, team- prefix handling) and computeEscalation() (security escalation, blocked cross-team, operations team from config, edge cases) - review-runner.test.mjs: add tests for normalizeSeverity() (uppercase, lowercase, mixed case, unknown values) - validate.test.mjs: add Phase 9 tests (valid dropdowns, invalid area, invalid severity, malformed template, area options with description separators) DRY improvements: - synchronize.mjs: add buildAreaRoutingTable() to generate routing from teams.yaml and expose as {{intakeAreaRoutingTable}} template variable - All 4 platform templates (Copilot, Cursor, Windsurf, Codex) now use {{intakeAreaRoutingTable}} instead of hardcoded routing strings Performance: - orchestrator.mjs: add loadTeamsSpec() cache (Map keyed by agentkitRoot) so resolveTeamByArea() and computeEscalation() read teams.yaml at most once per process. Export clearTeamsSpecCache() for test isolation. Robustness: - computeEscalation(): read operationsTeam from teams.yaml config instead of hardcoding 'team-quality' - issue-label-validation.yml: add workflow_dispatch trigger with optional issue_number input for manual dry-run testing All 597 tests pass. https://claude.ai/code/session_01Ue9aaiqqt1iHDx8KNwQ92m * chore: regenerate sync-backlog.prompt.md after template variable change The Copilot prompt template now uses {{intakeAreaRoutingTable}} which renders the area→team routing as a compact inline string instead of a full markdown table. https://claude.ai/code/session_01Ue9aaiqqt1iHDx8KNwQ92m * fix(branch-protection): address review findings for JSON commas, defaults, and Copilot API 1. Fix invalid JSON from {{#each}} blocks: add trailing commas to array items in bash templates and pipe through strip_trailing_commas helper 2. Fix securityAlertThreshold comment: correct valid values to none | critical | high_or_higher | medium_or_higher | all 3. Fix Copilot review API: use PUT /repos/{owner}/{repo}/copilot/code_review endpoint with review_new_pushes and review_draft_pull_requests in payload 4. Add bp* defaults in synchronize.mjs so missing branchProtection config produces valid scripts instead of broken JSON 5. Add 12 tests: branchProtection mapping coverage, array transform type, and JSON comma handling with resolveEachBlocks https://claude.ai/code/session_011Vp3WAEEQToMVkpdjNcimB * docs(issues): add issue specs for template update, finops agent, and team restructuring - cost-budget-flag-template-update: Document --budget flag in /cost template - finops-specialist-agent-consideration: Future split of FinOps from Data Engineer - agent-team-restructuring-gaps: Address docs/quality team gaps and orphaned agents https://claude.ai/code/session_01TdK27Dqh4JPrSqjCEUtCP2 * docs(issues): add agent/team analysis and gap issues - architect-agent-analysis: evaluate need for dedicated Architect agent - docs-quality-teams-missing-agents: T8 and T10 have no dedicated agents - data-agent-concerns-refactoring: separate cost analytics from Data Engineer - design-agents-missing-team: brand-guardian and ui-designer have no team - marketing-agents-missing-team: content-strategist and growth-analyst have no team - cost-budget-cli-flag-partial: --budget flag not exposed in command template - cost-review-handoff-chain-gap: no cost governance in any handoff chain - finops-md-content-verification: verify finops.md content across platforms https://claude.ai/code/session_01TdK27Dqh4JPrSqjCEUtCP2 * fix(branch-protection): apply PR #262 review comments - Replace {{#each}} in JSON heredocs with precomputed JSON vars (bpRequiredStatusChecksJson, bpCodeScanningToolsJson) to avoid invalid JSON from missing commas - Remove strip_trailing_commas helper (no longer needed) - Use $BRANCH/$Branch variable instead of hardcoded {{defaultBranch}} in code scanning and merge queue rulesets - Add branch name validation regex to prevent injection via --branch - Wrap code scanning and merge queue gh api calls with error handling (if ! ... in bash, try/catch in PowerShell) - Switch to unquoted heredocs (<<ENDJSON) where variable expansion is needed, keep quoted (<<'ENDJSON') where it is not - Use PowerShell here-strings (@"..."@) instead of hashtable literals for JSON payloads for consistency and correctness https://claude.ai/code/session_011Vp3WAEEQToMVkpdjNcimB * chore(sync): update generated file timestamps Sync engine updated last_updated timestamps from 2026-03-04 to 2026-03-05 across agents, chatmodes, and prompts. No content changes. https://claude.ai/code/session_01TdK27Dqh4JPrSqjCEUtCP2 * fix(template-utils): add JSON vars to RAW_TEMPLATE_VARS to preserve braces The sanitizeTemplateValue function strips `{` and `}` from string values in shell script templates. This caused precomputed JSON variables (bpRequiredStatusChecksJson, bpCodeScanningToolsJson) to lose their object braces, producing invalid JSON in generated scripts. Adding these vars to RAW_TEMPLATE_VARS bypasses sanitization since they contain pre-validated JSON that must preserve its structure. https://claude.ai/code/session_011Vp3WAEEQToMVkpdjNcimB * chore: regenerate files via agentkit sync https://claude.ai/code/session_011Vp3WAEEQToMVkpdjNcimB * docs(issues): add phased remediation plan for budget-guard review findings Addresses findings across security (command injection in shell hook), correctness (falsy-zero bug, regex ambiguity), performance (O(N) scan), and test coverage gaps. Organised into 4 phases by priority. https://claude.ai/code/session_01TdK27Dqh4JPrSqjCEUtCP2 * feat: add document-history command for institutional memory (#236) * feat(docs): improve template adoption across agent workflows and commands The docs/history/ TEMPLATE-* files existed but were rarely used because nothing in the agent workflow triggered their creation. This change adds history document creation prompts at every natural touchpoint: - CLAUDE.md template: add History Documentation section and /document-history to quick reference and standard session flow - team-TEMPLATE.md: add Step 4b (Create History Document) to team workflow and History Document section to output format - orchestrate.md: add history doc creation step to Phase 5 (Ship) - handoff.md: add history doc creation as output destination step 3 - review.md: add history doc recommendation to review output, update retrospective to use create-doc.sh for issue/lesson types - stop-build-check.sh: add non-blocking warning when significant source changes lack history documentation - New /document-history command: dedicated slash command with auto-detection of type and title from git history - commands.yaml: add document-history command spec - create-doc.sh: extend to support issue and lesson types - .index.json: add issue and lesson sequence keys - docs/history/README.md: document all six subdirectories and /document-history https://claude.ai/code/session_01VpYZ9Gr9JiMBKVJKKRujtD * fix(hooks): prevent grep -c exit code 1 from terminating stop hook grep -c returns exit code 1 when count is 0, which under set -euo pipefail would terminate the hook prematurely. Wrap in subshell with || true to handle zero-match case safely. Also fix table alignment in CLAUDE.md quick reference. https://claude.ai/code/session_01VpYZ9Gr9JiMBKVJKKRujtD * fix: address review findings across templates, hooks, and scripts - commands.yaml: expand --type enum to include issue/lesson with clarifying description that they're typically from retrospective - document-history.md: remove Bash(cat *) from allowed-tools (use Read instead), add merge-base fallback for shallow clones/short branches - stop-build-check.sh: hoist BRANCH/DEFAULT_BRANCH resolution to shared block to avoid redundant git calls, normalize indentation after refactor - create-doc.ps1: add issue/lesson support (ValidateSet, SubdirMap, initial sequences, sed placeholders, changelog skip logic) to match create-doc.sh parity https://claude.ai/code/session_01VpYZ9Gr9JiMBKVJKKRujtD * fix: address inline review findings for document-history, handoff, team template, and create-doc - Add Read/Write/Edit/Glob/Grep to document-history allowed-tools - Replace fragile hardcoded git ranges with commit-count-safe logic - Add deduplication guard to handoff history doc section - Rewrite team-TEMPLATE Step 4b to use /document-history --auto - Add issue/lesson types to create-doc.sh index initialization https://claude.ai/code/session_01VpYZ9Gr9JiMBKVJKKRujtD --------- Co-authored-by: Claude <noreply@anthropic.com> * feat(language-profile): add configurable inference modes and diagnostics (#239) * fix(claude-hooks): make session-start cross-platform and detect gh * feat(language-profile): add configurable inference modes and diagnostics * chore(sync): add split-pr helpers and telemetry workflow docs * fix(scripts): report created vs skipped quality issues * fix(sync): harden prereq checks and argument handling * docs(issue-170): track source-of-truth patch tasks and verification (#256) * fix(claude-hooks): make session-start cross-platform and detect gh * feat(language-profile): add configurable inference modes and diagnostics * chore(sync): add split-pr helpers and telemetry workflow docs * fix(scripts): report created vs skipped quality issues * fix(sync): harden prereq checks and argument handling * docs(issue-170): add source-file verification and task links * feat: remove idempotent issue creation script and streamline sync scripts - Deleted the `create-quality-issues.sh` script which was responsible for batch-creating GitHub issues for the quality framework expansion plan. - Updated `sync-split-pr.ps1` and `sync-split-pr.sh` scripts to remove redundant command assertions and improve logging output. - Simplified the usage instructions in `sync-split-pr.sh` and removed the command requirement checks, focusing on a cleaner execution flow. * docs(prd): implement autoupdate feature spec and CLI delivery milestone tracker (#244) * Initial plan * docs(prd): add autoupdate PRD, milestone tracker, and cross-references for CLI delivery issues Co-authored-by: JustAGhosT <5531814+JustAGhosT@users.noreply.github.com> --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: JustAGhosT <5531814+JustAGhosT@users.noreply.github.com> * feat: add expansion analyst agent with gap analysis engine (#264) * docs: add feasibility analysis for feature expansion agent Comprehensive analysis covering architecture fit, integration strategy, human-in-the-loop controls, spec generation approach, risk assessment, and phased implementation plan for a proposed expansion/addition agent. https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic * feat(expansion): add Phase 1 expansion analyst agent with analysis engine - ADR-08: Document decision to introduce expansion analyst agent - Add expansion-analyzer.mjs with 6 gap analyzers (docs, ADR, testing, security, ops, architecture) and scoring/ranking engine - Add suggestion-store.mjs with YAML persistence, lifecycle management, and rejection memory for deduplication - Add expansion-analyst agent definition to agents.yaml (product category) - Add /expand command to commands.yaml (read-only, no file writes) - Add comprehensive tests for both modules (all 606 tests passing) Phase 1 is analysis-only: produces scored suggestions as structured output without writing files or creating tasks. Human approval gates are required before any downstream action. https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic * Add phase gates and stopping conditions for expansion agent rollout Define explicit go/no-go criteria for each phase transition (analysis → suggestion store → spec generation → full integration) with measurable thresholds for signal-to-noise ratio, approval rates, draft quality, and performance. Include stopping conditions to abandon further phases if the agent produces more noise than value. https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic * Address PR #264 review comments: security, correctness, and cleanup - Fix path traversal vulnerability: validate suggestion IDs against strict pattern (SUG-NNN) before filesystem access (comments #4, #9) - Fix ID collision risk: namespace suggestion files by run timestamp to prevent overwrites across analysis runs (comment #3) - Add Write to /expand allowed-tools for --save flag (comment #1) - Change /expand from workflow to utility type to match docs (comment #2) - Derive operations docs path from docsSpec instead of hardcoding (comment #5) - Use nullish coalescing (??) for maxSuggestions/minImpact (comment #6) - Remove unused adrCount variable (comment #7) - Update header comment to reflect file-backed storage (comment #8) - Remove unused imports: basename, join, stat (comment #10) - Update tests for namespaced IDs and path traversal validation https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic * Regenerate outputs after /expand type change (agentkit sync) Run agentkit sync to propagate the /expand command type change from workflow to utility across all generated AI tool configurations. https://claude.ai/code/session_01WDZpGwDtTp9fADCJsLzXic --------- Co-authored-by: Claude <noreply@anthropic.com> * feat: dd issue sync script and ADR-08 for local-to-GitHub issue promotion (#263) * feat(issues): add gh fallback workflow, sync script, and ADR-08 When `gh` CLI is unavailable (proxy, air-gap, missing install), issues are recorded locally via `create-doc.sh issue`. A new `sync-issues.sh` script promotes unsynced local issue docs to GitHub Issues once access is restored (dry-run by default, --apply to create). - Update issue READMEs and templates with preferred/fallback workflow - Add Sync Status metadata section to issue template (gh_synced flag) - Add scripts/sync-issues.sh for local-to-GitHub issue sync - Add ADR-08 documenting sync trigger strategy (manual + CI) https://claude.ai/code/session_012h3YEeccLbQTMNUw1YDWHG * fix(sync-issues): address review findings for portability and robustness - Add --label missing-arg bounds check (prevents set -u crash) - Replace sed -i with portable temp-file pattern (macOS compat) - Redirect gh stderr to /dev/null, capture only stdout URL - Add comment explaining glob pattern naming convention assumption - Rename SYNCED counter to CANDIDATES for dry-run clarity - Fix ADR-08 branch reference from main to master (actual default) https://claude.ai/code/session_012h3YEeccLbQTMNUw1YDWHG * fix(sync-issues): address PR #263 review comments - Defer gh CLI preflight checks to --apply mode so dry-run works in air-gapped/restricted environments without gh installed - Make grep for H1 title non-fatal (|| true) so one malformed issue doc doesn't abort the entire sync batch under set -e - Update ADR README template source with ADR-07 and ADR-08 entries and regenerate via agentkit:sync to keep template and output in sync Previously fixed (prior commit): - --label missing-arg bounds check - sed -i replaced with portable temp-file pattern https://claude.ai/code/session_012h3YEeccLbQTMNUw1YDWHG * fix(sync-issues): address second round of PR #263 review comments - Capture gh stderr to temp file; surface it on failure instead of discarding with 2>/dev/null for better diagnostics - Make ISSUE_NUMBER extraction non-fatal using parameter expansion with grep fallback and "unknown" default, preventing set -e exit after a successful gh issue create - Metadata stamping (sed -> tmpfile -> mv) is now unconditional after gh issue create succeeds, preventing duplicate issues on next run - Fix default branch reference in ADR-08 from master to dev (actual default branch per git remote) - Update contradictory "None are implemented yet" wording to reflect Phase 1 (manual sync) is complete; only CI/hook triggers remain Note: history README template fix (contradictory create-doc wording) is blocked by the protect-templates hook — requires a manual edit to .agentkit/templates/docs/history/README.md followed by agentkit:sync. https://claude.ai/code/session_012h3YEeccLbQTMNUw1YDWHG --------- Co-authored-by: Claude <noreply@anthropic.com> * ci(sync): Standardize Node.js version management across CI workflows (#235) * ci: improve workflow naming and replace hardcoded node versions - Rename CI workflow from "CI" to "AgentKit Forge CI" for clarity - Add descriptive job names (Test, Validate, YAML Lint, Markdown Lint) - Remove node-version from matrix to eliminate meaningless "24" in job names (was showing as "test (24, windows-latest)" → now "Test (windows-latest)") - Replace hardcoded node-version: 24 with node-version-file: package.json in our own workflows (reads from engines.node in package.json) - Replace hardcoded node-version: 24 with node-version: lts/* in generated template files for consumer projects - Fix block-agentkit-changes.yml naming from snake_case to Title Case - Update branch-protection.yml comments to reflect new CI job names https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * ci: replace hardcoded CI values with template engine variables Add new spec fields (process.protectedBranches, documentation.historyPath, stack.nodeVersion, stack.pythonVersion) and corresponding template var mappings so CI workflow templates derive values from project.yaml instead of hardcoding them. - Gate audit-rust/audit-python jobs with {{#if hasLanguageRust/Python}} - Gate coverage-rust job with {{#if hasLanguageRust}} - Gate breaking-change version-file patterns by language booleans - Replace hardcoded branches: [main, dev] with {{protectedBranches}} - Replace hardcoded node-version: lts/* with {{nodeVersion}} - Replace hardcoded python-version: '3.12' with {{pythonVersion}} - Replace hardcoded docs/history/ paths with {{docsHistoryPath}} - Add missing GENERATED headers to documentation-quality.yml and documentation-validation.yml for consistency https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * ci: use testingCoverage from project.yaml in coverage-report threshold The coverage-report workflow previously required a manual COVERAGE_THRESHOLD env var to enforce thresholds. Now it defaults to {{testingCoverage}} from project.yaml (currently 80%) with env var override still supported. https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * ci: add template hygiene checks, mapping coverage validation, and always-on placeholder warnings - Add checkTemplateHygiene() to doctor.mjs with 5 detection rules for hardcoded values that should use template vars (nodeVersion, pythonVersion, protectedBranches, docsHistoryPath, testingCoverage) - Add validateMappingCoverage() to spec-validator.mjs to verify all PROJECT_MAPPING src paths exist in project.yaml - Promote unresolved placeholder warnings in template-utils.mjs from DEBUG-only to always-on for better visibility during sync - Gate dependency-audit.yml trigger paths with language conditionals (Cargo.toml/lock, pyproject.toml, requirements*.txt, *.csproj) - Add 8 new tests covering all new functionality https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * fix: address review findings for validateMappingCoverage and unused imports - Fix logic bug: `!resolved && current === undefined` → `!resolved || current === undefined` to catch leaf-level missing fields (e.g. stack.missingField when stack exists) - Remove unused imports: `relative` from path, `flattenProjectYaml` from template-utils - Remove redundant comment in spec-validator exports block - Add test for leaf-undefined edge case https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * chore(ci): regenerate sync outputs after spec validation fixes Sync removed dependency-audit trigger paths for languages not configured in project.yaml (Rust, Python, C#). The {{teamAccepts}} placeholder warning remains — requires engine-level fix in synchronize.mjs (tracked separately). https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * chore: regenerate sync outputs after dev merge Sync outputs updated from merge with dev branch including language-profile feature and document-history command. https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * fix(ci): address PR review comments and resolve doctor/sync bugs - Guard validateMappingCoverage return in doctor.mjs against undefined - Change file-level placeholder skip to per-line check in checkTemplateHygiene - Add validateMappingCoverage mocks to 4 language profile doctor tests - Remove dead DEBUG beforeEach/afterEach hooks in template-utils tests - Add teamAccepts and teamHandoffChain to all 4 teamVars blocks in sync - Pin Node.js version via .node-version file instead of package.json range - Update branch-protection script contexts to match renamed workflow - Wire COVERAGE_THRESHOLD from vars context in coverage-report workflow - Restore if-guard and wire vars in coverage-report template - Fix inline conditional YAML syntax in dependency-audit template - Widen .csproj glob to **/*.csproj for monorepo support - Add fallback defaults for nodeVersion, pythonVersion, docsHistoryPath Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(ci): use .node-version in dependency-audit workflow Replace lts/* with .node-version file for deterministic Node.js version. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * chore: regenerate sync outputs after review fixes Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com> * feat(governance): consolidate autoupdate docs and upstream-link guardrails (#257) * ci: improve workflow naming and replace hardcoded node versions - Rename CI workflow from "CI" to "AgentKit Forge CI" for clarity - Add descriptive job names (Test, Validate, YAML Lint, Markdown Lint) - Remove node-version from matrix to eliminate meaningless "24" in job names (was showing as "test (24, windows-latest)" → now "Test (windows-latest)") - Replace hardcoded node-version: 24 with node-version-file: package.json in our own workflows (reads from engines.node in package.json) - Replace hardcoded node-version: 24 with node-version: lts/* in generated template files for consumer projects - Fix block-agentkit-changes.yml naming from snake_case to Title Case - Update branch-protection.yml comments to reflect new CI job names https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * ci: replace hardcoded CI values with template engine variables Add new spec fields (process.protectedBranches, documentation.historyPath, stack.nodeVersion, stack.pythonVersion) and corresponding template var mappings so CI workflow templates derive values from project.yaml instead of hardcoding them. - Gate audit-rust/audit-python jobs with {{#if hasLanguageRust/Python}} - Gate coverage-rust job with {{#if hasLanguageRust}} - Gate breaking-change version-file patterns by language booleans - Replace hardcoded branches: [main, dev] with {{protectedBranches}} - Replace hardcoded node-version: lts/* with {{nodeVersion}} - Replace hardcoded python-version: '3.12' with {{pythonVersion}} - Replace hardcoded docs/history/ paths with {{docsHistoryPath}} - Add missing GENERATED headers to documentation-quality.yml and documentation-validation.yml for consistency https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * ci: use testingCoverage from project.yaml in coverage-report threshold The coverage-report workflow previously required a manual COVERAGE_THRESHOLD env var to enforce thresholds. Now it defaults to {{testingCoverage}} from project.yaml (currently 80%) with env var override still supported. https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * ci: add template hygiene checks, mapping coverage validation, and always-on placeholder warnings - Add checkTemplateHygiene() to doctor.mjs with 5 detection rules for hardcoded values that should use template vars (nodeVersion, pythonVersion, protectedBranches, docsHistoryPath, testingCoverage) - Add validateMappingCoverage() to spec-validator.mjs to verify all PROJECT_MAPPING src paths exist in project.yaml - Promote unresolved placeholder warnings in template-utils.mjs from DEBUG-only to always-on for better visibility during sync - Gate dependency-audit.yml trigger paths with language conditionals (Cargo.toml/lock, pyproject.toml, requirements*.txt, *.csproj) - Add 8 new tests covering all new functionality https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * fix: address review findings for validateMappingCoverage and unused imports - Fix logic bug: `!resolved && current === undefined` → `!resolved || current === undefined` to catch leaf-level missing fields (e.g. stack.missingField when stack exists) - Remove unused imports: `relative` from path, `flattenProjectYaml` from template-utils - Remove redundant comment in spec-validator exports block - Add test for leaf-undefined edge case https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * Initial plan * docs(prd): add autoupdate PRD, milestone tracker, and cross-references for CLI delivery issues Co-authored-by: JustAGhosT <5531814+JustAGhosT@users.noreply.github.com> * docs(reference): update delivery milestone tracker with issue links * docs(reference): format delivery milestone issue table * fix(docs): remove leftover merge markers in milestone tracker * chore(ci): regenerate sync outputs after spec validation fixes Sync removed dependency-audit trigger paths for languages not configured in project.yaml (Rust, Python, C#). The {{teamAccepts}} placeholder warning remains — requires engine-level fix in synchronize.mjs (tracked separately). https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * chore: regenerate sync outputs after dev merge Sync outputs updated from merge with dev branch including language-profile feature and document-history command. https://claude.ai/code/session_01HQwG1sTW7FyrbidZceZkQb * docs(autoupdate): address PR #257 review comments * Update last_updated dates for agents and prompts to 2026-03-05 * Initial plan * fix(docs): update dependency map to reference #258 consistently Address remaining Copilot review comment from PR #257 — replace "autoupdate issue (TBD)" with "#258 (autoupdate)" in the dependency map. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * feat(commands): add --branch and --project flags, surface flags in skill templates - Add --branch flag to review and preflight commands for explicit branch targeting - Add --project flag to review, check, healthcheck, and deploy for monorepo scoping - Add {{commandFlags}} section to all generic skill/command templates (Claude, Codex, Copilot, Cursor, Windsurf) so flags are visible at invocation time - Revert accidental blank line in doctor.mjs (protected file) - Regenerate all Copilot prompt outputs with flag tables Flag overlap analysis identified --stack (7 commands), --focus (5), --output (5), --dry-run (4), --scope (4) as the most shared flags. --branch and --project were entirely missing from the spec despite being natural scoping dimensions. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(cli): wire new spec flags into VALID_FLAGS and FLAG_TYPES Add --branch, --project, --open-issues, and --dry-run to the CLI runtime allowlist for check, review, healthcheck, and preflight commands so the parser accepts flags defined in commands.yaml. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: JustAGhosT <5531814+JustAGhosT@users.noreply.github.com> * feat(governance): add task governance controls with turn limits, loop detection, stagnation (#238) * fix: add intra-task loop prevention guards for agent execution Addresses the GPT 5.3 Copilot-style implementation loop where an agent repeatedly attempts the same action without progress. Adds four new guardrails to the team execution template: - MAX_TASK_TURNS (25): hard cap on agentic turns per task, escalates to input-required when exceeded - Action Repetition Detection: tracks last 5 actions, triggers on 3+ similar consecutive attempts - Stagnation Detection: flags tasks with no measurable progress after 10 turns - Handoff chain depth promoted from soft warning to hard stop Registers four new event types in orchestrate.md: TURN_LIMIT_REACHED, LOOP_DETECTED, STAGNATION_DETECTED, HANDOFF_DEPTH_EXCEEDED https://claude.ai/code/session_01APLR2oHKfwhra46MAWS41f * feat: make loop prevention limits configurable via teams.yaml Wire teamAccepts, teamHandoffChain, and three new loop prevention variables (maxTaskTurns, maxHandoffChainDepth, maxStagnationTurns) into the template rendering pipeline across all platforms (Claude, Cursor, Windsurf, Copilot). Teams can now override defaults per-team in teams.yaml: - max-task-turns: 25 (default) - max-handoff-chain-depth: 5 (default) - max-stagnation-turns: 10 (default) Also fixes two pre-existing bugs where teamAccepts and teamHandoffChain were referenced in team-TEMPLATE.md but never populated during sync, rendering as unresolved placeholders. Updates spec-validator to accept the new optional fields. All 576 tests pass. https://claude.ai/code/session_01APLR2oHKfwhra46MAWS41f * docs: add template variable audit and heuristic defaults plan Comprehensive analysis of all YAML spec → template variable wiring, identifying bugs (name mismatches, unwired vars), heuristic improvement opportunities, and open PR overlap. https://claude.ai/code/session_01APLR2oHKfwhra46MAWS41f * feat: fix unwired template variables, add heuristic defaults, and pipe syntax Phase 1 — Fix unwired/misnamed template variables: - loggingRetentionDays: templates used this name but mapping produced logRetentionDays; add dual mapping so both work - containerRuntime: add spec field (deployment.containerRuntime) and project-mapping entry - drTestSchedule: add spec field (compliance.disasterRecovery.testSchedule) and project-mapping entry Phase 2 — Heuristic defaults based on project context: - maxTaskTurns inferred from teamSize (solo=15, small=25, medium/large=35) - maxHandoffChainDepth inferred from team count (≤3→3, ≤6→5, >6→7) - maxStagnationTurns inferred from projectPhase (greenfield=15, active=10, maintenance/legacy=5) - testingCoverage fallback from projectPhase (greenfield=60, active=80, maintenance/legacy=90) Phase 3 — Template engine {{var|default}} pipe syntax: - Adds support for inline fallback values: {{testingCoverage|80}} - Resolves after normal substitution, before unresolved warning - Prevents literal {{varName}} rendering without requiring code changes - 3 new tests covering fallback, override, and empty default cases All 579 tests pass. https://claude.ai/code/session_01APLR2oHKfwhra46MAWS41f * docs: add functional + technical specs for commitConvention auto-detect and spec-defaults.yaml Two new feature specs to plan upcoming work: - SPEC-commit-convention-auto-detect: Extends agentkit discover to detect commit conventions from filesystem artifacts (.commitlintrc, .releaserc, package.json keys) and git-log heuristics. - SPEC-spec-defaults-yaml: Centralises template variable defaults into a declarative YAML file with static, phase-conditional, and teamSize-conditional blocks. Replaces scattered inline defaults across synchronize.mjs. Both specs include functional requirements, technical implementation details, testing strategy, and rollout plans. https://claude.ai/code/session_01APLR2oHKfwhra46MAWS41f * fix(governance): address PR review findings for task governance controls - Fix spec-validator schema: use `min` instead of `minimum` to match validator - Extract `buildTeamVars()` helper to eliminate 4x duplicated team variable blocks - Fix invalid `blocked` status references to use canonical `input-required` - Fix stagnation detection to track turn count (`_lastProgressTurn`) not timestamp - Remove duplicate separator line in synchronize.mjs - Sanitize pipe syntax values in template-utils.mjs for shell target safety - Update PLAN doc: reflect pipe syntax support, fix variable status, move to docs/ - Regenerate sync outputs Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude <noreply@anthropic.com> * fix(ci): add workflow to fix stuck branch protection checks Branch protection required status checks use names that don't match the actual CI check names (e.g. `markdown-lint` vs `Markdown Lint`), causing checks to stay in "Waiting for status" forever. This adds a manually-triggered workflow that: 1. Creates a tracking issue documenting the mismatch 2. Attempts to update branch protection check names via API https://claude.ai/code/session_01Q3u76p7dbyt2qzX62aaf4P * feat(editor): add brand tier and scheme config for IDE theme generation Add two new options to editor-theme.yaml: - `tier` (full/medium/minimal) — controls how many editor surfaces get branded. Minimal themes only the title bar for instant project identification; medium adds activity bar, status bar, and sidebar; full themes everything. - `scheme` (dark/light) — controls which color mapping wins when mode is 'both', so dark-scheme users get deep muted hues by default. The filterByTier() function in brand-resolver.mjs classifies VS Code color keys by surface prefix and strips non-matching slots before writing settings. Sync engine passes tier/scheme through to metadata. https://claude.ai/code/session_01QW2CpB4HDCWL8JES6N6Grq * fix(docs): add language specifier to fenced code block in issue template Fixes MD040 markdown lint failure in docs/history/issues/TEMPLATE-issue.md by adding `text` language to the fenced code block. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(branch-protection): address CodeRabbit review findings - Add defensive guards for bpCodeScanningTools entries in synchronize.mjs (filter nulls/scalars before property access) - Fix dry-run mode in setup-branch-protection scripts to skip label creation when --dry-run is set (both bash and PowerShell) - Fix PowerShell $Branch: scope qualifier issue by using ${Branch} - Rebuild docs/history/.index.json to reflect actual entry files Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(sync): address remaining CodeRabbit review findings - Set IaC toolchain to [terraform, terragrunt] in project.yaml to match template content that extensively references these tools - Fix coverage bullet indentation in dotnet/python language templates (was incorrectly nested under mocking bullet) - Fix ADR-02 template: replace placeholder date with {{syncDate}}, fill in repo-specific values for agentkit-forge instance - Fix marketing template: reference brand token name instead of hardcoding hex value (contradicted the "no hardcoded hex" rule) - Regenerated all outputs via agentkit sync --overwrite Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(branch-protection): harden error handling and input validation - Replace PowerShell try/catch with $LASTEXITCODE checks for native gh api calls (try/catch doesn't catch native command failures) - Filter out code scanning tools with empty names to prevent invalid API payloads ("tool": "" rejected by GitHub) - Filter bpRequiredStatusChecks entries to strings before JSON.stringify - Use jq `first` to handle duplicate rulesets returning multiple IDs - Remove redundant projectVars fallback (vars already contains merged values) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix(branch-protection): extract precomputation, add tests, fix docs - Extract buildBranchProtectionJson() into exported testable function - Add 13 unit …
JustAGhosT
added a commit
that referenced
this pull request
Mar 11, 2026
…ine (#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>
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
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.
➕ What does this PR do?
🔨 Changes
✅ Checklist
🗒 Notes for reviewer