Conversation
… filters, and context-mode integration - Redesign Console dashboard: 8 clickable stat cards (Projects, Sessions, Active, Memories, Extensions, Requirements, Specifications, Changes), Recent Specifications and Recent Memories cards - Flatten sidebar navigation: single nav list with inline ProjectFilter dropdowns per view - Add TopbarSpecs: active spec pills in top bar across all projects - Add notification navigation with project switching and clear-all support - Add session resume workflow: copy session ID, /resume deep-linking - Memory cards: show linked session, click to navigate, remove clutter - Cross-project APIs: /api/plans/active/all, /api/extensions?all=true - Project list from project_roots table (filtered to real git repos) - Update branding: Make Claude Code production-ready - Add context-mode integration for token-optimized context management - Update README, Docusaurus docs, and marketing site BREAKING CHANGE: Console UI completely restructured — sidebar flattened, project selector removed from sidebar and replaced with per-view inline filters, dashboard cards redesigned, PlanStatus and GitStatus widgets removed
…s, background jobs, and optional Telegram
# [8.0.0](v7.11.4...v8.0.0) (2026-04-08) ### Bug Fixes * pin Bun to 1.3.11 (1.2.15 lacks compression APIs, latest is unstable) ([2372ea7](2372ea7)) ### Features * add Pilot Bot — persistent automation agent with scheduled tasks, background jobs, and optional Telegram ([cab87f6](cab87f6)) * complete Console overhaul — v8.0.0 ([#126](#126)) ([bfa261f](bfa261f)) ### BREAKING CHANGES * Console UI completely restructured — sidebar flattened, project selector removed from sidebar and replaced with per-view inline filters, dashboard cards redesigned, PlanStatus and GitStatus widgets removed * fix: remove duplicate Help slide, update hooks count to 21 * fix: pin Bun to 1.2.15 in CI (latest has execSync regression on Linux)
## [8.0.1](v8.0.0...v8.0.1) (2026-04-08) ### Bug Fixes * use correct CronCreate parameter name (cron, not schedule) in bot skills ([ad7f4af](ad7f4af))
…down, session filtering, spec tab consistency
## [8.0.2](v8.0.1...v8.0.2) (2026-04-09) ### Bug Fixes * console dashboard overhaul — 2x2 recent cards, usage model breakdown, session filtering, spec tab consistency ([3648a2d](3648a2d))
## [8.0.3](v8.0.2...v8.0.3) (2026-04-09) ### Bug Fixes * Improved Code Reviewer Stability, Codegraph Usage and Context Mode ([e78d68b](e78d68b))
## [8.0.4](v8.0.3...v8.0.4) (2026-04-09) ### Bug Fixes * Improved Dependency Installation Speed for Installer and Updater ([e426d3c](e426d3c)) * Improved PRD and Spec Sharing / Annotation UI on pilot-shell.com ([afc32f7](afc32f7))
## [8.0.5](v8.0.4...v8.0.5) (2026-04-09) ### Bug Fixes * skip CodeGraph indexing in non-git directories ([b4281df](b4281df))
## [8.0.6](v8.0.5...v8.0.6) (2026-04-09) ### Bug Fixes * walk up directory tree for git repo detection in CodeGraph guard ([f028b57](f028b57))
## [8.0.7](v8.0.6...v8.0.7) (2026-04-10) ### Bug Fixes * add tool output compression + sandboxed content cards to Usage view ([46127a1](46127a1)) * hook venv sync, console annotation writes, codegraph native sqlite ([4e9d264](4e9d264)) * restore dom globals after terminal-preview-xss test to prevent portal ssr leak ([77eda81](77eda81)) * stop incomplete child_process mocks poisoning CI test runs ([7e30074](7e30074))
…ier, fix auto-mode flag and docs - Add chrome-devtools-mcp plugin installation to installer (marketplace + parallel install) - Update browser automation from 3-tier to 4-tier: Chrome → Chrome DevTools MCP → playwright-cli → agent-browser - Update rules (browser-automation, verification, testing), skills (spec-verify, spec-bugfix-verify, spec-implement) - Update docs: Docusaurus (open-source-tools, installation, mcp-servers, rules, spec workflow), site, README - Remove deprecated --enable-auto-mode flag from wrapper (fixes user-reported launch errors) - Fix project filtering to support git repo subdirectories in Console - Update hooks documentation count and footer links
## [8.0.10](v8.0.9...v8.0.10) (2026-04-14) ### Bug Fixes * codegraph native SQLite repair, /spec new-branch option, session prompt display ([21fb7b6](21fb7b6))
…anced UI - Add SessionJsonlService to parse Claude Code JSONL session files - Extract token usage, tool breakdown, model distribution, duration, cost - Cost calculation with up-to-date pricing for Opus 4.5+/4, Sonnet, Haiku tiers - New GET /api/sessions/:id/stats and POST /api/sessions/costs endpoints - Session detail modal with stats grid, prompt, tool usage (2-col layout) - Session cards show cost and total messages instead of observations - Sortable session list columns (date, cost, messages, prompts) - Dashboard shows cost per session, cross-project navigation - 15 unit tests with JSONL fixtures
# [8.1.0](v8.0.10...v8.1.0) (2026-04-15) ### Features * add session details with JSONL stats, cost calculation, and enhanced UI ([0590783](0590783))
- Add customization packs: pilot customize install/update/remove/status CLI commands for teams on Team/Enterprise tiers to overlay custom skills, rules, hooks, and agents on top of core Pilot via a git repository. Auto-reapplied on pilot update and installer runs. Security: symlink rejection, license validation, atomic manifest writes, stale file cleanup on re-apply. - Add Opus 4.7 model support across launcher, console, and documentation. - Support custom plan statuses in Console and statusline — teams can write arbitrary Status: values in plan files and they display as-is with neutral styling instead of being hidden. - Never skip Codex adversarial review in spec-plan/spec-verify/spec-bugfix-plan — wait for the task completion notification, never read the output file to determine completion. - Docs: new customization page, updated pricing, extensions cross-ref, README.
# [8.2.0](v8.1.0...v8.2.0) (2026-04-16) ### Features * customization packs, Opus 4.7 support, and Codex bugfixes ([af711a4](af711a4))
…rtifacts untracked
## [8.2.1](v8.2.0...v8.2.1) (2026-04-17) ### Bug Fixes * customization overrides, skill decomposition polish, generated artifacts untracked ([ec1c268](ec1c268)) * Removed static effort for skills/commands to adjust based on global ([4fd9050](4fd9050))
## [8.2.2](v8.2.1...v8.2.2) (2026-04-17) ### Bug Fixes * Updated inconsistent steps in spec workflow ([a1e6118](a1e6118))
## [8.2.3](v8.2.2...v8.2.3) (2026-04-17) ### Bug Fixes * drop skill banner, normalize step heading levels across workflow skills ([05d37fa](05d37fa))
…n can't leak workers Fixes #129. ## Problem The SessionEnd hook does two side-effects: 1. POST session-complete to Console (localhost:41777) 2. Stop the bun worker-service.cjs if this is the last active session Both currently run synchronously in the hook process: - `_complete_session` calls `urllib.request.urlopen(..., timeout=5)` inline - `main` calls `subprocess.run(["bun", ...], timeout=15)` inline When Claude Code's harness cancels the SessionEnd hook (which it does after a short internal deadline), the hook process is terminated before either side-effect completes. Symptoms: - Terminal shows "SessionEnd hook [...] failed: Hook cancelled" - bun worker-service.cjs daemons accumulate as orphans across sessions - Orphans hold open file handles to ~/.pilot/memory/pilot-memory.db (WAL) but serve no traffic (only the newest is bound to :41777) Reproduced on Arch/Manjaro with Claude Code 2.x — observed three running daemons (two orphaned) after a handful of sessions. ## Fix Hand both side-effects off to detached subprocesses (`start_new_session=True`) before returning from `main`. The parent hook becomes essentially instantaneous (two `Popen` calls, no waiting), so harness cancellation no longer races the work. Ordering: worker-stop first, Console POST second. The worker-stop is the leakier side effect (holds file descriptors + network port); prioritising it means that even in a pathological case where the second `Popen` fails, the critical resource release still happens. Timeouts inside the detached workers are preserved (Console POST 5s, no explicit timeout on `bun stop` — unchanged from previous behaviour because it now runs in its own session and doesn't tie up the hook). The Console POST logic is inlined as a string constant (`_COMPLETE_SESSION_WORKER`) passed to `python -c`. This avoids creating a new file in the repo and keeps the worker self-contained; the subprocess startup cost (~30ms) is the only overhead when Console is down. ## Verification - `python -m py_compile` on the modified file: passes - `main()` with no env vars: returns 0 cleanly (short-circuits on missing CLAUDE_PLUGIN_ROOT) - Inlined worker script compiles and executes cleanly when invoked with a live Console (tested with `python3 -c <script> <url> <sid>` → POST 200) - Behavioural smoke test: run a session, exit, observe no "Hook cancelled" message and no orphan `bun worker-service.cjs --daemon` processes in `ps` - Data safety: workers are stateless HTTP fronts over shared SQLite-WAL; killing the old synchronous path's leaked workers with SIGTERM had no data-integrity impact (PRAGMA integrity_check = ok) ## Why not just increase the hook timeout? The hook timeout is controlled by the harness, not this repo. Even if it were raised, network stalls / bun startup latency / loaded-disk fsync spikes can all push a synchronous implementation past any fixed deadline. Detaching removes the class of failure entirely.
…coped annotations Bundles several independent fixes under one release. ## Vector DB disk exhaustion (closes #130) hnswlib's persisted `link_lists.bin` could grow to 11.5 TB logical / 859 GB physical, filling a 1 TB WSL disk. Adds three independent layers so the file cannot grow unbounded: - `VectorDbSizeGuard` module with both logical (`st.size`) and physical (`st.blocks * 512`) measurement so sparse-file bloat is detected. - Boot-time guard in `DatabaseManager` evicts the dir before Chroma starts (defaults: 2 GB physical / 50 GB logical, configurable via `CLAUDE_PILOT_VECTOR_DB_MAX_PHYSICAL_MB` / `CLAUDE_PILOT_VECTOR_DB_MAX_LOGICAL_MB`). - Pre-write gate in `ChromaSync.assertSafeToWrite()` — every `chroma_add_documents` / `chroma_create_collection` call checks size first (tighter caps: 1 GB physical / 10 GB logical). hnswlib never receives a write that would extend an already-oversized file. - Post-retention guard in `RetentionScheduler` re-checks after every cleanup and closes Chroma if tripped. - Auto-disable after 2 evictions within 7 days atomically flips `CLAUDE_PILOT_CHROMA_ENABLED=false` in `settings.json`. Search falls back to SQLite via existing `SearchOrchestrator` path (same escape hatch as claude-mem #991). - `/api/vector-db/health` now reports physical size, largest file, and last-eviction marker so operators can see what happened. ## Project-scoped annotations in Changes view Annotations were visible across projects — two concurrent "Changes" sessions showed each other's annotations on the same screen. Scopes `AnnotationRoutes` queries by project and filters the Changes view accordingly. ## Model routing settings UI Surface per-role model selection (opus/sonnet/haiku) in the console Settings view. Launcher `model_config.py` + `settings_injector.py` read `~/.pilot/config.json` and inject model preferences into plugin settings on startup. ## Spec workflow refinements - spec-bugfix-plan: expanded red-flags step, tightened root-cause plan + write-plan steps. - spec-bugfix-verify: behaviour-contract audit in verify-fix and plan-verify steps. - spec-implement: TDD-loop guidance updated. - docs: workflows/spec.md and features/{console,model-routing, statusline}.md updated to match. ## Statusline widgets Widget formatter updates with corresponding test coverage. ## Installer `claude_files.py` gains ~216 lines for non-destructive settings/skill merges on re-install; tests updated accordingly. ## Bundled assets `pilot/scripts/*.cjs` and `pilot/ui/*.js` regenerated from `console/src/` to match the fixes above. ## Verification - Bun: 1349 pass / 0 fail (22 new tests for `VectorDbSizeGuard`) - Python: 1574 pass / 0 fail - typecheck clean
## [8.2.4](v8.2.3...v8.2.4) (2026-04-21) ### Bug Fixes * **hooks:** make SessionEnd fully non-blocking so harness cancellation can't leak workers ([3cb965a](3cb965a)) * prevent vector-db disk exhaustion + model routing UI + project-scoped annotations ([1a066d7](1a066d7)), closes [#130](#130) [#991](https://github.com/maxritter/pilot-shell/issues/991)
…e task cards - statusline: read rate_limits from Claude Code stdin (cross-platform, no OAuth), single-percentage pacing arrow, fall back to lines+git when the field is absent (API/Enterprise users). Drop token-count suffix on the context bar. - installer: vendor installer/skill_builder.py and build SKILL.md in-process during install; no longer shells out to the pilot binary. Drift-guard test keeps it byte-identical to launcher/skill_builder.py. Bump parallel download workers 8 -> 16. - launcher/wrapper: generic capability probe caches claude --help once and appends --thinking-display summarized when supported. - launcher/codegraph: silence redundant native-SQLite repair message. - console/Topbar: spec pills now show hover cards with parsed task lists (completed + next pending) and bugfix/feature icons. - console/Changes: prefer main-branch plans over worktree plans when auto-selecting active plan so annotations match the current checkout; CodeReviewPanel filters diffs by path. - skills: prd/03-clarify skips questions already answered upstream; spec-bugfix-plan gains the git log -S pattern for token-change forensics; verify steps tightened. - settings: enable showThinkingHint. - ignore local claude-pace/ reference clone.
# [8.3.0](v8.2.4...v8.3.0) (2026-04-23) ### Features * cross-platform statusline usage, in-process skill build, console task cards ([9f394be](9f394be))
Primary feature: /benchmark — quantitative before/after measurement for Claude Code skills and rules. Isolated per-run sandboxes, crash-proof global-rule auto-hiding, parallel workers, and an LLM grader that emits structured per-assertion verdicts plus improvement suggestions. - pilot/skills/benchmark/: orchestrator + 6 step files (intake → target discovery → author evals → execute → present findings → improvement plan), runner.py with parallel workers and per-run filesystem isolation, aggregate_benchmark.py, isolation.py, progress.py, utils.py, agents/grader.md, and ~1800 LOC of unit test coverage - docs/docusaurus/docs/workflows/benchmark.md: workflow documentation Also in this commit: - /prd skill: new 02b-ideate step, orchestrator and 8 steps updated - console: pilot-spawn utility + tests, license routes tests, user-message handler fixes, worker-service refactor, Dashboard/Requirements/Topbar viewer updates - installer: finalize step + claude_files test updates - launcher: banner and statusline formatter updates - CI/release: release-dev and release workflow tweaks, pre-commit hook - docs/site: ConsoleSection, InstallSection, WorkflowSteps, PRD and spec workflow pages, intro, statusline feature page - misc: README, .gitignore, pyproject.toml
# [8.4.0](v8.3.0...v8.4.0) (2026-04-24) ### Features * add /benchmark skill and evaluation framework ([843ef85](843ef85))
console/usage:
- Replace external ccusage CLI with native Claude session analytics
- New pricing engine (LiteLLM fetch + bundled fallback + model aliases)
filters out <synthetic>/unknown internal markers
- New /api/usage/{daily,monthly,models,yield,plan} endpoints; cross-view
parity ensures Sessions and Usage agree on cost
- Restructured Usage view: equal-height pairs (Cost-by-Model + Routing,
Code-Quality + Tool-Output-Compression), Plan progress below
- Cost-by-Model groups by family (Opus/Sonnet/Haiku) with distinct colors
- Code-Quality card: Day/Week/Month toggle, commits-shipped headline,
cost-per-shipped-commit, productive-spend %, degraded-state surfacing
- Plan tracking via SettingsRoutes (Pro/Max/Custom + month-end forecast)
- Yield uses execFile + per-call timeouts + ref validation + HEAD-sha cache
ccusage scrub:
- Drop install_ccusage from installer + tests, uninstall hint, and the
open-source-tools docs row; codeburn/ source repo deleted
pilot/hooks:
- New tool_redirect hook (with tests) to redirect deferred sub-agent flows
pilot/rules:
- New documentation-sync rule; tweaks to development-practices, testing,
and verification
pilot/skills:
- benchmark: runner/utils tightening, review-iterate + improvement-plan
step rewrites, new test coverage
- setup-rules: split out new sync-agents step, renumber summary 11 -> 12
docs:
- Refreshed benchmark, setup-rules, statusline, rules, intro pages
- Site SEO/sitemap/indexnow plumbing, hero copy, Schema.org logo fix
- Stddev/pass-rate disambiguation and removal of phantom Step 6 refs in
benchmark.md (CodeRabbit findings)
## [8.4.1](v8.4.0...v8.4.1) (2026-04-27) ### Bug Fixes * native usage analytics, ccusage scrub, hook + rules + site updates ([c7cd604](c7cd604))
Adds /fix as the standard bugfix command — investigate, RED test, fix, audit, done — with mandatory end-to-end verification. Slims the existing /spec bugfix lane substantially. The workflow change addresses the issue where two simple bugs took 21+ minutes and 400k tokens halfway through. ## /fix — new user-invocable command - New skill at pilot/skills/fix/ with six steps (investigate → RED → fix → audit → quality → finalise). Always quick: stops cleanly and tells the user to re-invoke with /spec when complexity exceeds the quick lane. - Iron Law #4: never claim done from unit tests alone. Step 4.3 is mandatory end-to-end verification — UI bugs must run browser automation via Claude Code Chrome / Chrome DevTools MCP / playwright-cli / agent-browser; CLI/API/library bugs must be exercised with real invocations and concrete evidence captured in the completion report. - Step 1.4 adds explicit Instrument-when-needed guidance with SPEC-DEBUG markers (cleanup grep enforced at audit). - Registered as user-invocable, model: opus. ## Slimmed full bugfix lane (/spec) - Dropped Codex from spec-bugfix-plan entirely (was step 7) — adversarial review on a plan with no code yet was overhead. Codex still runs at spec-verify time. - Codex-once rule across spec-plan + spec-verify: a session-scoped sentinel file prevents re-runs across plan iterations within one /spec invocation (was re-running on every annotate/verify cycle). - spec-bugfix-verify: 11 steps → 8. Dropped redundant plan-verify (no-op), runtime-verification (foldable into quality), post-merge-verify (squash merges with no conflicts are safe). - spec-bugfix-verify Step 2 (Behavior Contract audit): tightened from ~6.4KB to ~3.8KB. Sub-steps 2.5/2.6 merged into 2.4. Step 2.6 now enforces mandatory E2E verification with concrete evidence — no claim-VERIFIED-on-tests-alone. - spec-bugfix-plan: tightened red-flags step with Common Rationalizations and User Signals tables (inspired by superpowers/systematic-debugging). Step 3.3 adds a concrete multi-layer instrumentation example. Step 4 fix-approach-selection defaults to picking the obvious fix rather than manufacturing fake alternatives. - spec-implement Task 2 (bugfix lane): drops the per-task full-suite run. For bundled bugfix plans this was the single biggest token sink — full suite now runs once at the Quality Gate, targeted module test runs between fix iterations. ## Iteration cap - spec-bugfix-verify Step 8 caps fix iterations at 3. After three failed verify cycles, AskUserQuestion presents Continue/Pivot/Abandon — stops the fix-on-fix-on-fix death spiral that previously had no bound. ## Dispatcher / model defaults / Console - /spec dispatcher restored to clean split: /spec routes bugs to spec-bugfix-plan (full lane), /fix is a separate user-facing command for the quick lane. No --full flag, no auto-routing between commands. - launcher/model_config.py: /fix and /benchmark added to default skill models (opus). Console useSettings.ts mirrors. - Console Settings page: General Rows now lists /fix and /benchmark with descriptive labels + sub-text. SPEC_PHASE_ROWS rendered with sub-text. - launcher/banner.py + installer/steps/finalize.py: /fix and /benchmark added to quick-start tips and post-install workflow list. Spec-Driven feature description updated to mention both /spec and /fix. - pilot/settings.json: companyAnnouncements line now lists /fix; spinner tip updated. ## Docs - New Docusaurus page docs/workflows/fix.md with workflow diagram, Common Issues table, mandatory E2E section, /spec vs /fix decision table. Sidebars updated. - /spec page trimmed of bugfix-only content (now points at /fix). - README: new /fix section with How /fix works + How /spec handles bugs details blocks. - Marketing site WhatsInside.tsx updated to mention /fix. ## Other-session changes incorporated - Console: parsePlanContent.ts + tests for spec section rendering and plan annotator persistence (Bug A + Bug B fixes). - pilot/hooks/tool_redirect.py + tests: extended. - pilot/.mcp.json + hooks/hooks.json + scripts/*.cjs: minor updates. - pilot/rules/development-practices.md + mcp-servers.md: updates. - installer/downloads.py: minor. - pilot/ui/* compiled bundles: rebuilt. - .devcontainer/*: minor adjustments.
|
The latest updates on your projects. Learn more about Vercel for GitHub. 1 Skipped Deployment
|
|
Caution Review failedPull request was closed or merged during review WalkthroughThis PR represents version 8.4.1, repositioning Pilot Shell as a "Claude Code Engineering Platform" with major documentation overhaul, new workflows ( Changes
Estimated code review effort🎯 4 (Complex) | ⏱️ ~75 minutes Possibly related PRs
Suggested labels
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
⚔️ Resolve merge conflicts
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. Review rate limit: 7/8 reviews remaining, refill in 7 minutes and 30 seconds.Comment |
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
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds /fix as the standard bugfix command — investigate, RED test, fix, audit, done — with mandatory end-to-end verification. Slims the existing /spec bugfix lane substantially. Addresses the issue where two simple bugs took 21+ minutes and 400k tokens halfway through.
/fix — new user-invocable command
pilot/skills/fix/with six steps (investigate → RED → fix → audit → quality → finalise)--fullflag, no auto-routing between commands — honour the user's command choice)SPEC-DEBUG:markers (cleanup grep enforced at audit)Slimmed full bugfix lane (/spec)
Iteration cap
spec-bugfix-verify Step 8 caps fix iterations at 3. After three failed verify cycles, AskUserQuestion presents Continue/Pivot/Abandon — stops the fix-on-fix-on-fix death spiral.
Dispatcher / model defaults / Console
Docs
docs/workflows/fix.mdwith workflow diagram, Common Issues table, mandatory E2E section, /spec vs /fix decision tableOther-session changes incorporated
parsePlanContent.ts+ tests for spec section rendering and plan annotator persistence (Bug A + Bug B fixes from a parallel session)Verification
tsc --noEmitcleanruff checkclean on all files I touchedSummary by CodeRabbit
Release Notes
New Features
/benchmarkworkflow: Quantitative before/after evaluation with graded assertions./fixworkflow: Streamlined bugfix process with RED→fix→verify discipline.Improvements