docs: fix stale file path references#4
Merged
Conversation
ajonkisz
added a commit
that referenced
this pull request
Mar 29, 2026
* Add markdown rendering for web_fetch preview - Create MarkdownWebContent custom element with preview/raw toggle - Add isMarkdownContent() detection (URL-based + content heuristics) - Conditionally render markdown-block for markdown content in WebFetchRenderer Co-Authored-By: Bobbit <noreply@bobbit.dev> * Add unit tests for isMarkdownContent detection heuristics - tests/fixtures/markdown-web-content.html: file:// fixture with inline isMarkdownContent function matching the design doc spec - tests/markdown-web-content.spec.ts: 17 Playwright tests covering: - URL-based detection (.md, .markdown, query params, hash, case-insensitive) - Content-based detection (headings+links, frontmatter, code fences, bold+lists, images) - False positive avoidance (plain text, single heading, HTML, JSON) - URL priority over content check Co-Authored-By: Bobbit <noreply@bobbit.dev> * Fix test fixture to match production isMarkdownContent Co-Authored-By: Bobbit <noreply@bobbit.dev> * Document markdown rendering in web_fetch tool previews Co-Authored-By: Bobbit <noreply@bobbit.dev> * Add unit tests for archived team agent filtering logic Tests verify the fixed filtering logic from renderGoalGroup(): - Mapped members appear under live team lead - Unmapped members (no/empty teamLeadSessionId) appear under live lead - Members belonging to archived leads stay with their lead - Delegates and team-lead roles are excluded - Orphan members (pointing to non-existent lead) appear under live lead - Unmapped fallback works when no archived leads exist - Members from other goals are excluded Co-Authored-By: Bobbit <noreply@bobbit.dev> * fix: show archived team agents in sidebar when 'See Archived' toggled Bug 1: Broaden archivedForLiveLead filter to include members with no teamLeadSessionId or a stale one not matching any archived lead. Previously required exact match on live lead ID, missing all unmapped members. Bug 2: Add fallback rendering for unmapped archived members when no archived leads exist. Previously .map() on empty archivedLeads silently dropped them. Co-Authored-By: Bobbit <noreply@bobbit.dev> * Fix duplicate rendering of unmapped archived members * docs: add debugging tip for archived session sidebar rendering Add a new entry to the Debugging tips section in AGENTS.md explaining how archived team members are mapped to team leads in the sidebar, including the teamLeadSessionId fallback logic in renderGoalGroup(). Co-Authored-By: Bobbit <noreply@bobbit.dev> * feat: add Claude Code memory bridge to system prompt assembly - Add parseMemoryFile() to parse YAML frontmatter from memory files - Add readClaudeMd() to read CLAUDE.md with @ref resolution and dedup - Add readClaudeCodeMemories() to read ~/.claude/projects/{cwd}/memory/*.md - Update assembleSystemPrompt() to inject CLAUDE.md and memories - Update getPromptSections() to expose new sections in inspector UI Filters memory files by type (user/feedback/project), caps at 20 files and 16000 chars total. Skips silently if directories don't exist. Co-Authored-By: Bobbit <noreply@bobbit.dev> * test: add unit tests for Claude Code Memory Bridge functions Tests for parseMemoryFile, readClaudeMd, and readClaudeCodeMemories. All tests expect 'not a function' until implementation is merged. Covers: valid/invalid frontmatter parsing, CLAUDE.md dedup logic, @ reference resolution, memory type filtering, 20-file cap, 16000-char content cap, MEMORY.md skip, missing directory handling. Co-Authored-By: Bobbit <noreply@bobbit.dev> * docs: add Claude Code memory integration section to AGENTS.md Co-Authored-By: Bobbit <noreply@bobbit.dev> * docs: update system prompt assembly in features.md, improve AGENTS.md memory section Co-Authored-By: Bobbit <noreply@bobbit.dev> --------- Co-authored-by: Artur Jonkisz <aj@Arturs-MacBook-Pro-M5-Max.local> Co-authored-by: Bobbit <noreply@bobbit.dev>
ajonkisz
added a commit
that referenced
this pull request
May 1, 2026
…lan_status Live-test report from agent-memory: when an architect ran as the LLM verify step for a `parent`-workflow plan-review gate's "DAG correctness" check, they hit an interactive permission prompt: > Role "Architect" doesn't have access to goal_plan_status The architect needs to fetch the proposed plan to assess its DAG correctness; same for the spec-auditor's "Spec completeness" check. Both are LLM-review verify steps on the canonical parent workflow (see `seed-default-workflows.ts` plan-review gate). Root cause: `Children` tool group default is `ask` (intentional, so arbitrary roles can't silently spawn child goals); only `team-lead` overrides at the group level. Reviewer roles inherit `ask` and get prompted at runtime, blocking the verify step until a human clicks through. Fix: per-tool override `goal_plan_status: allow` on architect AND spec-auditor. Only the read-only inspection tool is opened up. The five mutating tools (goal_spawn_child, goal_plan_propose, goal_merge_child, goal_pause, goal_resume) remain `ask` for these roles \u2014 the team-lead is still the only role that can mutate plan state without a human prompt. Resolution priority (see `tool-activation.ts::resolveGrantPolicy`): priority #1 (role tool-specific) beats priority #4 (group default), so the per-tool `allow` cleanly overrides the `Children: ask` group default for these roles only. Tests: tests/builtin-roles-children-policies.test.ts (NEW, 10 cases): - architect.toolPolicies.goal_plan_status === "allow" - resolveGrantPolicy("goal_plan_status", "Children", architect, ...) === "allow" with `Children: ask` group default - mutating Children tools (5) still resolve to "ask" for architect - same trio for spec-auditor - other reviewer roles (code-reviewer, security-reviewer, qa-tester) do NOT silently get the override (least-privilege circuit-breaker) - team-lead's group-level Children: allow still intact (sanity) Total node tests now 1557 (was 1547, +10 new). Verified: - npm run check → passes - npm run test:unit (Node) → 1557 / 1557 pass Co-authored-by: bobbit-ai <bobbit@bobbit.ai> Co-Authored-By: Bobbit (Claude Opus 4 7) <bobbit@bobbit.ai>
SuuBro
added a commit
that referenced
this pull request
May 4, 2026
Sweep documentation to match the steer-subsystem rewrite (commits f37aadd, 3d3d34c, 377f4bb, 6ed08fc): - docs/prompt-queue.md: drop dispatched-flag tracking; document shadow ledger (inFlightSteerTexts) lifecycle, _dispatchSteer single site, exactly-once force-kill recovery via _reconcileAfterAbort + enqueueAtFront. - docs/debugging.md: rewrite Abort/steer/queue entries — duplicate-on-Stop is the canonical pre-rewrite bug; new diagnostic order points at ledger splice / reconciliation paths. - docs/internals.md: update Steer-interruptible bash_bg wait call-site map (single _dispatchSteer abort site + steerQueued unblock site = 2 total), and the implicit-unstick paragraph that referenced the deleted dispatched: true flag. - docs/features.md: replace resetDispatched()+drainQueue() with shadow-ledger reconciliation in the force-abort recovery summary. - docs/testing-strategy.md / docs/testing-tier-2-5.md: drop STEER_RECEIVED references; document new mock contract (handlePrompt round-trip, transcript-only assertions). - AGENTS.md: add 'Modify steer / queue dispatch' recipe; rewrite the 'Live-steer lost / duplicated after Stop' debug entry; link the design doc from Reference docs. - docs/design/steer-subsystem-rewrite.md: status -> implemented; append '9. Implementation outcome' with commit map, what landed, verification greps, deviations from the original design (mitigation B used; pill retained; reconciliation order shifted). All four verification greps from the spec return zero (or, for grep #4, the expected positive count) on master. Co-authored-by: bobbit-ai <bobbit@bobbit.ai>
SuuBro
added a commit
that referenced
this pull request
May 4, 2026
…le (#473) * design: steer subsystem rewrite — single source of truth via SDK clearQueue() Co-authored-by: bobbit-ai <bobbit@bobbit.ai> * refactor(steer): remove dispatched flag, single dispatch site - Drop QueuedMessage.dispatched and the markDispatched/removeDispatched/ resetDispatched/dequeueUndispatched triplet from PromptQueue. - Add PromptQueue.enqueueAtFront() for reconciliation paths. - Replace _dispatchSteeredMessages with a single _dispatchSteer that removes rows from the queue *before* awaiting rpcClient.steer(); on RPC failure, rows are re-enqueued at the front via enqueueAtFront. - deliverLiveSteer happy path now routes through _dispatchSteer. - Tool-boundary and agent_end safety-net dispatchers pre-pop via dequeueAllSteered() and feed _dispatchSteer. - Drop the message_end(user) -> removeDispatched() hook and the resetDispatched() calls in agent_end(wasAborting) and forceAbort. Per docs/design/steer-subsystem-rewrite.md \u00a73. Co-authored-by: bobbit-ai <bobbit@bobbit.ai> * test(steer): drop STEER_RECEIVED mock ACK; assert on transcript/DOM Co-authored-by: bobbit-ai <bobbit@bobbit.ai> * test(steer): P0 reconnect/multitab/stop-exactly-once + queue unit tests Co-authored-by: bobbit-ai <bobbit@bobbit.ai> * Reapply "fix(review): mobile bottom-sheet popover + RP-09 submitted-flag race" This reverts commit 1289817. * fix(steer): shadow-ledger reconciliation on abort + AC §3 restart test Gap-analysis findings on the implementation gate: 1. SDK reconciliation on abort (design §6.1 mitigation B). The upstream pi-coding-agent RPC bridge cannot be extended to proxy AgentSession.clearQueue(), so we mirror the SDK's text-match splice logic at the Bobbit layer for the abort window only: - inFlightSteerTexts ledger on SessionInfo, populated AFTER rpcClient.steer() resolves in _dispatchSteer. - _consumeSteerEcho() splices the entry on the matching message_end(role:user), mirroring agent-session.js:265-280. - _reconcileAfterAbort() drains the ledger and re-enqueues at front as steered, called from agent_end while wasAborting and from forceAbort after rpcClient.stop(). This closes the live-steer-then-abort window where the SDK accepted a steer but the bridge died before its message_end echoed. 2. AC §3 — gateway-restart durability test. New tests/e2e/steer-gateway-restart.spec.ts simulates restart in the in-process harness via SessionManager.restoreSessions(). Asserts both steered messages dispatch exactly once and ordering is preserved across persisted-messageQueue → restoreSession → drainQueue. 3. Minor nits: - tests/e2e/ui/bg-wait-steer-stop-flow.spec.ts wraps body in try/finally with deleteSession cleanup. - tests/e2e/steer-multitab.spec.ts uses STAY_BUSY:3000 to match the goal spec verbatim. Co-authored-by: bobbit-ai <bobbit@bobbit.ai> * docs(steer): update for shadow-ledger architecture Sweep documentation to match the steer-subsystem rewrite (commits f37aadd, 3d3d34c, 377f4bb, 6ed08fc): - docs/prompt-queue.md: drop dispatched-flag tracking; document shadow ledger (inFlightSteerTexts) lifecycle, _dispatchSteer single site, exactly-once force-kill recovery via _reconcileAfterAbort + enqueueAtFront. - docs/debugging.md: rewrite Abort/steer/queue entries — duplicate-on-Stop is the canonical pre-rewrite bug; new diagnostic order points at ledger splice / reconciliation paths. - docs/internals.md: update Steer-interruptible bash_bg wait call-site map (single _dispatchSteer abort site + steerQueued unblock site = 2 total), and the implicit-unstick paragraph that referenced the deleted dispatched: true flag. - docs/features.md: replace resetDispatched()+drainQueue() with shadow-ledger reconciliation in the force-abort recovery summary. - docs/testing-strategy.md / docs/testing-tier-2-5.md: drop STEER_RECEIVED references; document new mock contract (handlePrompt round-trip, transcript-only assertions). - AGENTS.md: add 'Modify steer / queue dispatch' recipe; rewrite the 'Live-steer lost / duplicated after Stop' debug entry; link the design doc from Reference docs. - docs/design/steer-subsystem-rewrite.md: status -> implemented; append '9. Implementation outcome' with commit map, what landed, verification greps, deviations from the original design (mitigation B used; pill retained; reconciliation order shifted). All four verification greps from the spec return zero (or, for grep #4, the expected positive count) on master. Co-authored-by: bobbit-ai <bobbit@bobbit.ai> * docs(steer): correct _reconcileAfterAbort placement in forceAbort It runs immediately after rpcClient.stop() (post-kill), not before. The ledger is Bobbit-owned in-process state so the timing is correct either way, but the docs claimed pre-kill placement which is factually wrong and could mislead someone debugging the abort path. Co-authored-by: bobbit-ai <bobbit@bobbit.ai> --------- Co-authored-by: Josh S. <8768403+SuuBro@users.noreply.github.com> Co-authored-by: bobbit-ai <bobbit@bobbit.ai>
ajonkisz
added a commit
that referenced
this pull request
May 13, 2026
Findings addressed (all required, per review): - #1 Wire project config keys - read pre_warm_lsp, lsp_max_servers, lsp_idle_ttl_ms, lsp_disabled from ProjectConfigStore at supervisor construction. Kill switch returns lsp_unavailable from every tool. - #2 Cold-start progress signalling - new GET /api/lsp/state probe plus 500ms delayed status frame from the tool extension when state is starting/cold. - #3 Crash detection - typescript client invokes supervisor onClose callback; supervisor drops dead entries; 3 crashes / 60s trips a 5-minute cooldown. crashState map persists across entry deletions. - #4 acquire() wired - session-setup acquires after preWarm so refcount reflects attached sessions; release rearms idle timer; acquire after release cancels armed timer. - #5 tsconfig/package.json file-watch - fs.watch the worktree dir with 1500ms debounce, graceful shutdown + lazy respawn. - #6 Sandbox bridge plumbing - MultiProjectSandboxLspBridge constructed once SandboxManager is wired via setSandboxBridge(). Removed the sandboxed gate from session-setup pre-warm sites. Full container path round-tripping for outbound results deferred (noted in code). - #7 Path containment in dispatch - reject absolute and dot-dot paths with LspUnavailableError. Tests cover both. - #8 Removed fixed 300ms sleeps in typescript-client.spec.ts; verified non-flaky across 5 runs. Tests: +7 supervisor specs (crash x2, acquire-cancels-timer, config watch, disabled, preWarm disabled, path containment x2). All 19 tests under tests/lsp/ pass. Tool-description budget passes (lsp counted). Co-authored-by: bobbit-ai <bobbit@bobbit.ai>
SuuBro
added a commit
that referenced
this pull request
May 18, 2026
* fix(staff): persist staffId on session spawn + one-shot backfill migration Staff sessions lost their staffId association across server restart / in-place respawn because StaffManager set `session.staffId = id` only in memory \u2014 SessionManager.createSession() never accepted `staffId` in its opts, so plan.staffId stayed undefined and persistOnce wrote no staffId to disk. On respawn, restoreSession read ps.staffId = undefined \u2192 no BOBBIT_STAFF_ID env \u2192 defaults/tools/inbox/extension.ts gated all three inbox tools off \u2192 staff agent could no longer mark entries complete. Fix: - Thread staffId through createSession opts \u2192 plan \u2192 persistOnce, so it lands in PersistedSession at spawn time and survives any respawn path. - Add a one-shot startup backfill (src/server/agent/staff-backfill.ts) that heals existing broken sessions by matching title + worktreePath against the staff registry, with a loud warn log on every match (spec req #4). Wired from server.ts boot via sessionManager.backfillStaffIds(staffManager). Regression test: tests/staff-session-staffid-persistence.test.ts \u2014 Tests 1 and 3 fail on master, pass after fix. Source guards pin the read side (restoreSession), the write side (persistOnce), and the new plan-builder wiring (createSession opts \u2192 plan). Co-authored-by: bobbit-ai <bobbit@bobbit.ai> * docs: document staffId-persistence fix + add debugging entry Co-authored-by: bobbit-ai <bobbit@bobbit.ai> --------- Co-authored-by: Josh S. <8768403+SuuBro@users.noreply.github.com> Co-authored-by: bobbit-ai <bobbit@bobbit.ai>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes 10 stale path references across 3 doc files:
workflows/→.bobbit/config/workflows/config/system-prompt.md→.bobbit/config/system-prompt.md;tests/playwright-e2e.config.ts→playwright-e2e.config.ts(root)extensions/,roles/,workflows/) to include correct.bobbit/prefix