* fix(gsd): enforce backtick file paths in task plan IO sections
The reactive task graph (ADR-004) derives dependencies from backtick-wrapped
file paths in ## Inputs and ## Expected Output sections. Without concrete
paths, the graph is ambiguous and falls back to sequential execution.
Changes:
- task-plan.md template: add comments explaining paths are machine-parsed
- plan-slice.md prompt: explicitly instruct planner to write backtick file
paths in IO sections, add self-audit check for path presence
- observability-validator.ts: new validation rules missing_output_file_paths
(warning) and missing_input_file_paths (info) catch plans without paths
- plan-quality-validator.test.ts: 4 new test cases for IO path validation
* fix(ci): increase max_tokens and add JSON parse error handling in ai-triage
max_tokens: 300 was too low, causing truncated JSON responses from Claude
that failed to parse. Bumped to 1024 and added try/catch with raw text
logging for easier debugging.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add reactive (graph-derived parallel) task execution within slices.
When enabled via preferences, the dispatch table derives a task dependency
graph from IO annotations in task plans and dispatches multiple ready,
non-conflicting tasks in parallel via subagent.
Architecture:
- Graph derivation happens at dispatch time (auto-dispatch.ts)
- A new reactive-execute prompt instructs the agent to use subagent
parallel mode to dispatch all currently-ready tasks
- The auto-loop treats reactive-execute as a single unit type
- After agent_end, the orchestrator checks which tasks completed and loops
New files:
- reactive-graph.ts: pure graph derivation, ready-set resolution,
conflict detection, deadlock detection, IO loader, state persistence
- prompts/reactive-execute.md: prompt template for parallel dispatch
- tests/reactive-graph.test.ts: 22 unit tests for graph functions
- tests/reactive-executor.test.ts: 11 integration tests for dispatch
rules, preferences validation, state persistence, re-entry
Modified files:
- types.ts: TaskIO, DerivedTaskNode, ReactiveExecutionConfig,
ReactiveExecutionState interfaces
- files.ts: parseTaskPlanIO() extracts IO from task plan sections
- preferences-types.ts: reactive_execution config + known keys
- preferences-validation.ts: validation with range checks
- auto-dispatch.ts: new reactive-execute dispatch rule
- auto-prompts.ts: buildReactiveExecutePrompt()
- auto-recovery.ts: artifact verification for reactive-execute
- auto-post-unit.ts: reactive state cleanup on slice completion
Backward compatible: disabled by default, falls through to sequential
execution when disabled, ambiguous, or only 1 task is ready.
* feat: add anthropic-vertex provider for Claude models on Google Vertex AI
Add a new anthropic-vertex provider that enables using Claude models
(Opus 4.6, Sonnet 4.6, Haiku 4.5) through Google Vertex AI using the
@anthropic-ai/vertex-sdk package. Follows the same pattern as the
existing google/google-vertex provider split.
Detection uses ANTHROPIC_VERTEX_PROJECT_ID (same env var as Claude Code)
with CLOUD_ML_REGION for region selection, falling back to us-central1.
Extracts shared Anthropic utilities into anthropic-shared.ts (message
conversion, tool conversion, param building, stream processing) to
avoid duplication between anthropic.ts and anthropic-vertex.ts.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add full Claude model set for anthropic-vertex provider
Add 200K context window variants for Opus 4.6 and Sonnet 4.6, plus
older models (Sonnet 4.5, Sonnet 4, Opus 4.5, Opus 4.1, Opus 4, Haiku 4.5).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: add @anthropic-ai/vertex-sdk to root dependencies
Required for the published package to resolve the vertex SDK at runtime.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: remove unnecessary comments to match codebase style
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: remove duplicate stream functions after rebase
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Nathan Roe <nathan.roe@carvana.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Replace the simple notifyRemoteAutoActive notification with an interactive
guardRemoteSession menu that shows session details and offers actionable
choices (view status, steer, stop, or force start). Guards all auto-mode
entry points: bare /gsd, /gsd next, and /gsd auto.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(#1526): auto-mode worktree commits land on main instead of milestone branch
GitServiceImpl.getMainBranch() was designed to detect manual /worktree worktrees
(worktree/<name> branches) but incorrectly applied the same logic to auto-mode
worktrees (milestone/<MID> branches). When no worktree/<name> branch existed,
it fell back to the current branch, which in certain contexts could be main,
causing slice commits to land on main instead of the milestone branch.
Fix: Detect if currently on a milestone/* branch first (auto-mode case) and
return it, before checking for worktree/* branches (manual worktree case).
- Modify getMainBranch() to detect milestone branches first
- Add test verifying getMainBranch() returns correct branch in auto-worktree
- All tests pass, build succeeds
Fixes#1526
* fix: cmux library directory incorrectly loaded as extension
The extension auto-discovery in resolveExtensionEntries() finds
index.js files in subdirectories and treats them as extensions.
The cmux directory has an index.js but it's a utility library
(imported by gsd and subagent extensions), not an extension itself.
Two changes:
1. When a package.json has a "pi" manifest, treat it as authoritative
and don't fall through to index.ts/index.js auto-detection. This
lets library directories opt out with "pi": {}.
2. Add package.json to cmux directory with empty pi manifest.
Report generation in auto-loop uses native dynamic import() which
bypasses jiti's alias resolution. The import chain
metrics.js → mod.js → ui.js → @gsd/pi-tui failed because Node
cannot resolve @gsd/pi-tui from ~/.gsd/agent/extensions/.
Split ANSI-aware layout helpers (padRight, joinColumns, centerLine,
fitColumns) into layout-utils.ts and keep format-utils.ts pure so
report modules can import formatting functions without pulling in
the @gsd/pi-tui dependency.
GitServiceImpl.getMainBranch() was designed to detect manual /worktree worktrees
(worktree/<name> branches) but incorrectly applied the same logic to auto-mode
worktrees (milestone/<MID> branches). When no worktree/<name> branch existed,
it fell back to the current branch, which in certain contexts could be main,
causing slice commits to land on main instead of the milestone branch.
Fix: Detect if currently on a milestone/* branch first (auto-mode case) and
return it, before checking for worktree/* branches (manual worktree case).
- Modify getMainBranch() to detect milestone branches first
- Add test verifying getMainBranch() returns correct branch in auto-worktree
- All tests pass, build succeeds
Fixes#1526
The Anthropic API's max_uses resets per request — when pause_turn triggers
a resubmit, the model gets a fresh budget each time. This allowed unlimited
total searches across a research unit, overwhelming the TUI render buffer.
Fix:
- Count web_search_tool_result blocks in conversation history on each
before_provider_request to track cumulative searches per session
- Cap total native searches at 15 per session (3 full turns of 5)
- Dynamically set max_uses to min(5, remaining) — preserves per-turn cap
while enforcing session ceiling
- When budget exhausted, omit web_search tool entirely instead of letting
the model hit max_uses_exceeded repeatedly
- Reset counter on session_start (new agent unit)
- Add web search budget guidance to research prompts (defense in depth)
Tests: 5 new tests covering budget tracking, exhaustion, and reset.
All 35 native-search tests pass.
* feat(dashboard): two-column layout with redesigned widget
- Two-column layout: progress bar left, task checklist right
- 4 widget modes: full → small → min → off (cycle with /gsd widget)
- Health indicator and ETA in header line for immediate visibility
- Simplified stats: 3 items (hit rate, cost, context %) instead of 7
- Short PWD (last 2 segments), git worktree name with ⎇ prefix
- Last commit time + message in footer (cached every 15s)
- Preview script with mock data for all modes
* docs: add dashboard widget screenshots for PR #1530
* docs: update dashboard screenshots with wider renders
* docs: wider full-width dashboard screenshots
* feat(dashboard): persist widget_mode in preferences
- Add widget_mode to GSDPreferences and KNOWN_PREFERENCE_KEYS
- Load saved widget_mode from preferences on first access
- Persist to global PREFERENCES.md on /gsd widget change
- Default remains "full" when no preference is set
Remove the bundled SwiftUI skill which had 13+ broken references to a
non-existent `../macos-apps/references/` directory. Add a CI script
that validates all relative .md file references in bundled skills,
preventing this class of bug from shipping again. Fix 5 additional
pre-existing broken references in other skills.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When a unit fails and is retried, the model tier is now escalated
(light -> standard -> heavy) if dynamic routing is enabled and
escalate_on_failure is not explicitly disabled. This connects the
existing escalateTier() function, which was fully implemented and
tested but never called at runtime.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Bare /gsd and /gsd next now check for a remote auto-mode session via
readSessionLockData before attempting to start step-mode. If another
process holds the lock, a steering menu is shown instead of competing
for the lock and killing the running session.
Also fixes the guided-flow "all slices discussed" message to detect
active auto-mode and direct users to /gsd status instead of bare /gsd.
Closes#1507
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Wire resolveProfileDefaults into loadEffectiveGSDPreferences so
token_profile: budget actually sets cheaper models and skips phases
- Add /gsd rate <over|ok|under> command to submit user feedback on
model tier assignments, completing the adaptive routing feedback loop
- Document that models config is required for dynamic routing activation
- Document ceiling behavior when dynamic routing is active
- Document reassess_after_slice as required for reassessment
Closes#1505 (partial — escalateTier wiring deferred to follow-up)
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The onCompromised callback in the retry acquisition path lacked the
elapsed-time suppression that the primary path had, causing unconditional
_lockCompromised=true on benign mtime drift. Additionally, validateSessionLock
now attempts PID-based recovery and re-acquisition before declaring the lock
lost, preventing sessions from stopping when no other process has taken over.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace loadAgentInstructions() with a deprecation warning that fires when
legacy agent-instructions.md files are detected. Pi core already supports
AGENTS.md (with CLAUDE.md fallback) per directory, making the custom GSD
mechanism redundant.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Directories under .gsd/milestones/ that don't match the M\d+ pattern
(e.g. slices/, temp-backup/) are now excluded instead of being returned
with their raw name. This prevents rogue directories from blocking
auto-mode milestone discovery.
Closes#1494
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
isValidationTerminal() now normalizes verdict: passed → pass before
comparison. LLM-generated validation files that write "passed" instead
of "pass" are accepted as terminal, preventing milestones from being
treated as incomplete.
Closes#1429
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add missing `parseRoadmap` import in `auto-dispatch.ts`
- Add missing `unlinkSync` import in `auto.ts`
- Add missing `syncGsdStateToWorktree` import in `worktree-sync-milestones.test.ts`
All three were dropped during the PR #1419 merge.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* rfc: GitOps branching & versioning strategy proposal
Proposes a Git-Flow Lite model with automated integration branches:
main ← production-ready, tagged releases only
next ← integration branch for next minor (PRs target here)
release/X.Y ← stabilization branch, only bugfixes allowed
hotfix/X.Y.Z ← emergency fixes cherry-picked to release
Includes:
- RFC document with lifecycle diagrams, migration path, open questions
- Workflow scaffolds (in docs/proposals/workflows/, NOT .github/):
- create-release.yml: manual dispatch to cut release branch from next
- sync-next.yml: auto-sync next branch after version tags
- backmerge.yml: auto back-merge release fixes to next
This is an experimental proposal requesting community feedback before
any implementation. The workflow files are inert scaffolds — they do
not run in CI.
* fix: prevent ensureGitignore from adding .gsd when tracked in git (#1364)
CRITICAL DATA-LOSS FIX: ensureGitignore() unconditionally added '.gsd' to
.gitignore even when .gsd/ was a real git-tracked directory, causing git to
report ~889 tracked files as deleted.
Root cause: BASELINE_PATTERNS included '.gsd' unconditionally, and the
gitignore modification ran BEFORE migration checks in auto-start.ts.
Changes:
- Add hasGitTrackedGsdFiles() helper using nativeLsFiles to detect tracked
.gsd/ content
- ensureGitignore() now skips the '.gsd' pattern when .gsd/ has tracked files
- untrackRuntimeFiles() now skips entirely when .gsd/ has tracked files
- migrateToExternalState() aborts when .gsd/ has tracked files
- Reorder auto-start.ts: migration runs BEFORE gitignore modification
- Add 8 regression tests covering all scenarios
Fixes#1364
* fix: break recursive dialog loop when all milestones complete (#1348)
Two interacting bugs:
1. Recursive dialog loop: When all milestones are complete, bootstrapAutoSession
calls showSmartEntry → sets pendingAutoStart → checkAutoStartAfterDiscuss
calls startAuto → bootstrapAutoSession → showSmartEntry → infinite loop.
The discuss workflow completes without producing a milestone directory, so
phase stays 'complete' and the cycle never breaks.
Fix: Add a re-entry counter (_consecutiveCompleteBootstraps) that tracks
how many times bootstrapAutoSession enters the 'complete' branch without
advancing. After 2 consecutive attempts, break the loop with a warning
message and return false.
2. Missing _releaseFunction = null in retry lock onCompromised handler:
The retry lock path in session-lock.ts set _lockCompromised but didn't
null out _releaseFunction, which could leave a stale reference that
masks the compromise detection in validateSessionLock().
Fixes#1348
* fix: self-heal stale roadmap checkbox for interrupted complete-slice (#1350)
When complete-slice is interrupted after writing SUMMARY.md and UAT.md but
before flipping the roadmap checkbox, auto-mode enters an infinite loop —
re-launching the same complete-slice unit because the dispatch loop uses
the roadmap checkbox as the sole 'slice done' signal.
Fix: Add a self-heal case in selfHealRuntimeRecords that detects when
SUMMARY + UAT exist but the roadmap checkbox is unchecked, and auto-fixes
the checkbox. This allows the verification to pass and the dispatch loop
to advance.
Fixes#1350
* fix: add EISDIR guard to complete/validate milestone prompts (#1343)
The LLM was passing tasks/ directory paths to the read tool during
milestone completion, causing EISDIR crashes. Added file system safety
instructions to both complete-milestone and validate-milestone prompts
telling the LLM to use ls/find for directory listing, not the read tool.
Fixes#1343
* feat: improve extension conflict messages with removal guidance (#1347)
When a user extension registers tools/commands that now ship as built-ins,
the conflict message now includes '(built-in tool supersedes — consider
removing <path>)' and the log level is downgraded from 'Extension load error'
to 'Extension conflict'.
Changes:
- resource-loader.ts: detect built-in vs user extension conflicts, add hint
- cli.ts: downgrade severity for superseded-tool conflicts
Fixes#1347
* test: fix always-skipped preferences test, add test:marketplace script
- preferences.test.ts: Replace always-skipped getIsolationMode test with
a filesystem-independent version that validates the default through
validatePreferences() instead of reading ~/.gsd/preferences.md.
Reduces skipped count from 3 → 2.
- package.json: Add test:marketplace script for running marketplace
contract tests (claude-import-tui, plugin-importer-live,
marketplace-discovery) with GSD_TEST_CLONE_MARKETPLACES=1.
These tests need external repos and self-skip in unit test runs.
Remaining 2 skips:
- Marketplace contract test suites (need external repos, run via test:marketplace)
- Windows-only tests in validate-directory.test.ts are platform-conditional
and correctly skip on macOS
* fix: use execFileSync in regression tests for Windows portability
The regression tests used execSync with shell-dependent constructs:
- '&&' command chaining (works in bash/cmd but fragile)
- Single-quoted commit messages (bash-only, cmd.exe splits on spaces)
Replaced with execFileSync via a git() helper that bypasses the shell
entirely. Each git operation is a separate call with proper argument
arrays, eliminating all shell interpretation issues.
Fixes windows-portability CI failure.
* fix: guard milestone completion against missing slice summaries (#1368)
Auto-mode could report a milestone as complete after executing only the
last slice, skipping earlier unexecuted slices. The milestone completion
signal fired based on roadmap checkbox state, which could be stale or
inconsistent after worktree transitions.
Changes:
- auto-dispatch.ts: Added slice SUMMARY file existence check to both
validating-milestone and completing-milestone dispatch rules. If any
slice lacks a SUMMARY file, dispatch stops with a diagnostic error
instead of proceeding to validation/completion.
- validate-milestone.test.ts: Updated tests to create slice summary
files (required by the new guard).
- file-watcher.test.ts: Fixed flaky 'auth.json change emits auth-changed
event' test by adding watcher initialization delay and increasing event
propagation timeout (race condition when run in full suite).
Fixes#1368
* fix: warn on common misspelled preference keys + verify field guidance (#1373, #1341)
#1373: Users setting 'taskIsolation.mode: none' instead of 'git.isolation: none'
got a generic 'unknown key' warning. Added KEY_MIGRATION_HINTS that map common
misspellings (taskIsolation, task_isolation, isolation, manage_gitignore, auto_push,
main_branch) to their correct git.* equivalents with actionable messages.
#1341: Planning agent writes aspirational prose in Verify fields ('Sections 3.1
and 3.2 exist with exact formulas. Zero TBD.') instead of executable commands.
Added explicit verify field rules to the plan template: must be mechanically
executable, with examples of good vs bad patterns for content tasks.
Fixes#1373, partially addresses #1341
* refactor: extract roadmap-mutations.ts + shared test-utils.ts
Consolidation:
- roadmap-mutations.ts: Extracted markSliceDoneInRoadmap() and markTaskDoneInPlan()
from duplicated implementations in doctor.ts, mechanical-completion.ts, and
auto-recovery.ts. All three callers used identical regex patterns.
mechanical-completion.ts and auto-recovery.ts now import the shared utility.
(doctor.ts deferred — touched by PR #1349)
- test-utils.ts: Shared cross-platform test utilities for GSD extension tests.
Provides git() helper (execFileSync, no shell), makeTempRepo() with
core.autocrlf=false, cleanup(), createFile(), safeReadFile(), and
writeMilestoneFixture(). 12 test files currently define their own versions
of these helpers — new tests should import from test-utils.ts instead.
Security audit: No injection vectors (sid/tid are alphanumeric from roadmap
parser), no path traversal, no secrets, no new dependencies.
* fix: port conflict false positive on non-Node projects + paused worktree resume (#1381, #1383)
projects without package.json. macOS AirPlay Receiver listens on port 5000,
causing a spurious warning on non-Node projects.
Fix: Skip port checks entirely when no package.json exists. When using
default ports, filter out 5000 on macOS.
in-memory only. Re-entering /gsd started a fresh bootstrap from the project
root instead of the active worktree.
Fix: pauseAuto() now writes paused-session.json to .gsd/runtime/ with
milestoneId, worktreePath, originalBasePath, and stepMode. startAuto()
checks for this file before bootstrap and restores the paused session
context, including worktree re-entry. stopAuto() cleans up the file.
Fixes#1381, #1383
* fix: catch spawn ENOENT in uncaught exception guard + snapshot session lock path (#1384, #1363)
uncaught exception and crashes auto-mode. The EPIPE guard now also catches
ENOENT from spawn syscalls — logs the error and continues instead of
terminating the process.
the lock path differently via gsdRoot() because basePath could be either the
project root or a worktree path. gsdRoot() produces different results for
each, so the lock was written to one path and validated against another.
Fix: Snapshot the resolved lock path (_snapshotLockPath) at acquisition time
and reuse it for all subsequent lock operations within the session.
Fixes#1384, #1363
* fix: suppress false-positive lock compromise + skip migration with active worktrees (#1362, #1337)
because the event loop stall delays the heartbeat mtime update. The handler
now checks elapsed time since acquisition — if within the 30-minute stale
window, it logs a warning and continues instead of setting _lockCompromised.
Real takeovers (past the stale window) still trigger the compromise flag.
even when .gsd/worktrees/ contained active git worktrees with locked
directory handles. This caused EBUSY errors and destructive data loss.
Migration now checks for active worktree directories and skips entirely
if any are found.
Fixes#1362, #1337
In worktree isolation mode, the secrets gate checked for .env at the
worktree path (process.cwd()), but the user's .env lives at the project
root. Keys that existed in the project root's .env were reported as
missing, causing repeated blocking key collection prompts.
Fix: getManifestStatus() now accepts an optional projectRoot parameter.
When provided (worktree mode), it checks both the worktree .env AND the
project root .env. All callers in auto.ts and auto-start.ts updated to
pass s.originalBasePath.
Fixes#1387
In worktree isolation, process.cwd() drifts when async_bash or background
jobs change directory, causing commits to land on the wrong branch.
Realign cwd to basePath before every dispatch and hook dispatch.
Also clean stale .git/SQUASH_MSG, MERGE_HEAD, MERGE_MSG after failed
squash-merges to prevent subsequent git operations from seeing phantom
merge state.
Adapted to post-M001 architecture: cwd fix in auto-loop.ts runUnit(),
hook cwd fix in auto.ts dispatchHookUnit(), merge cleanup in
worktree-resolver.ts _mergeWorktreeMode().
Co-authored-by: Lex Christopherson <lex@glittercowboy.com>
syncGsdStateToWorktree() assumed the milestones/ directory already
existed in the worktree. On a fresh worktree bootstrap this directory
is absent, causing milestone sync to silently skip all entries and
auto-mode to report "All milestones complete" immediately.
Create the directory before iterating if the main repo has milestones
but the worktree does not.
Co-authored-by: Berat Can <berat@hyperlab.games>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Post-unit hooks that use browser tools can hang indefinitely when the LLM
calls browser_wait_for with condition 'network_idle' against a dev server
with persistent connections (Vite HMR, WebSocket). The networkidle event
never fires because at least one connection stays open.
Fix: Inject a browser safety instruction into every hook prompt warning
against network_idle and recommending selector_visible, text_visible, or
delay as alternatives.
Fixes#1345
migrateToExternalState() moved .gsd/ to ~/.gsd/projects/<hash>/ and
created a symlink, but never verified the symlink resolved correctly.
On Windows, junction creation can silently fail or resolve to the wrong
target. If the symlink was broken, the backup (.gsd.migrating) was
deleted anyway, losing all project state.
Changes:
- migrate-external.ts: After creating symlink, verify it resolves to the
expected path and is readable. If verification fails, restore from backup.
- repo-identity-worktree.test.ts: Canonicalize temp dirs with realpathSync
to fix macOS /var → /private/var mismatch in path assertions.
- resource-loader.ts: Check for agents/ subdir before using dist/resources
as source — partial builds (tsc without copy-resources) create an
incomplete dist/resources that's missing agents/ and skills/.
Fixes#1377
When the user's home directory is a git repo, resolveProjectRoot() correctly
returns $HOME as the main tree root. assertSafeDirectory() then hard-blocks it
with the home-directory guard added in #1053, even though the session is running
from a valid GSD worktree (e.g. ~/.gsd/worktrees/M001) — not from $HOME itself.
Fix: in projectRoot(), detect when CWD diverges from the resolved root (i.e. we
are inside a git worktree) and validate the CWD instead. The worktree path is
never $HOME, so the guard no longer fires. When not in a worktree cwd === root,
preserving the existing behaviour unchanged.
Adds a regression test: validateDirectory() on a ~/.gsd/worktrees/M001 path
must return { safe: true, severity: "ok" }.
Co-authored-by: Jeremy McSpadden <jeremy@fluxlabs.net>
* fix(gsd extension): detect initialized projects in health widget
Use .gsd presence plus project-state detection for the health widget so bootstrapped projects no longer appear as unloaded before metrics exist.
* fix(gsd extension): make health widget execution-aware
Lead the health widget with current GSD execution state so it explains what the project is doing before surfacing provider and environment diagnostics. Keep issue, budget, and progress details as secondary context and cover the new output with focused widget tests.
* fix(gsd extension): address review feedback on health widget PR
- Replace em dash with ASCII hyphen in headline for terminal safety
- Reformat catch/finally to standard single-line style
- Replace computeProgressScore() status with direct phase labels so
the status reflects the actual execution phase, not a global health
aggregate
- Use lightweight milestone-dir scan instead of full detectProjectState()
to avoid unnecessary filesystem work on the 60s refresh
- Add cache warm-up comment on updateSliceProgressCache call
- Add safety comment on early void refresh() call
- Update test assertions for new phase labels and ASCII separator
* fix: replace symlink-follow in gsdRoot() with git-root-anchored walk-up discovery
The old implementation blindly assumed .gsd lived at basePath and only
followed symlinks as a migration escape hatch. This caused the health
widget to show "No project loaded" when:
- .gsd was moved to a non-default location
- cwd was a subdirectory of the project root
- the session started inside a worktree path
New probe chain in gsdRoot():
1. basePath/.gsd — fast path (common case, zero overhead)
2. git rev-parse root — anchors to the repo root regardless of cwd
3. Walk up from basePath — finds .gsd in an ancestor (bounded by git root)
4. basePath/.gsd — creation fallback for init/new projects
Key correctness detail: basePath is normalized via realpathSync before
any comparisons, ensuring the git-root boundary check works on macOS
where /var is a symlink to /private/var. Walk-up only runs when inside
a git repo and only when basePath != gitRoot — preventing escape into
unrelated filesystem directories.
Result is cached per-basePath for the process lifetime. All 52 callers
of gsdRoot() benefit with no call-site changes.
Adds tests/paths.test.ts covering all 6 probe cases.
* fix: correct report() call signature in paths.test.ts — takes no arguments
* fix: normalize git output paths and use realpathSync.native for Windows compatibility
- Use path.normalize() on git rev-parse output to convert forward slashes
to backslashes on Windows, so the git-root boundary check fires correctly
- Use realpathSync.native() instead of realpathSync() to resolve both
symlinks (macOS /var→/private/var) and 8.3 short names (Windows RUNNER~1)
- Update test tmp() helper to use realpathSync.native so expected paths
match the resolved paths returned by probeGsdRoot
The fallback path for GSD-WORKFLOW.md still referenced the legacy .pi
directory (~/.pi/GSD-WORKFLOW.md) instead of the correct .gsd/agent
location. This broke workflow dispatch when GSD_WORKFLOW_PATH env var
was not set.
- Update fallback path from ~/.pi/ to ~/.gsd/agent/ in three call sites
(dispatchWorkflow, dispatchDoctorHeal, handleTriage)
- Sync GSD-WORKFLOW.md to agentDir during initResources() as a fallback
for alternative entry points that may not set the env var
Co-authored-by: Berat Can <berat@hyperlab.games>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add a new /gsd changelog command that fetches releases from the GitHub API,
filters by version, and sends the raw changelog into the conversation for the
LLM to summarize the most important changes.
- New changelog.ts module: GitHub API fetch, semver filtering, body parsing
- Routing block in commands.ts with lazy import (same pattern as forensics)
- Tab completion in commands-bootstrap.ts TOP_LEVEL_SUBCOMMANDS
- Help text under VISIBILITY section in showHelp()
- No new npm dependencies — uses built-in fetch()
* refactor: replace recursive auto-dispatch with linear autoLoop, delete ~3k lines of dead code
Replace the complex recursive dispatch system (dispatchNextUnit, reentrancy
guards, stall detection, idempotency tracking, skip-depth machinery) with a
simple linear while(s.active) loop in auto-loop.ts.
Key changes:
- New auto-loop.ts with autoLoop(), runUnit(), resolveAgentEnd()
- Deleted auto-idempotency.ts, auto-stuck-detection.ts, session-lock.ts,
mechanical-completion.ts, progress-score.ts, auto-constants.ts, unit-id.ts
- Extracted WorktreeResolver class for worktree path resolution
- Added auto-worktree-sync.ts for worktree synchronization
- Simplified auto.ts from ~1400 lines to ~400 lines
- Fixed 9 TypeScript errors (NotifyCtx type widening, capture typing)
- Comprehensive test coverage: 32 auto-loop tests + worktree resolver/DB tests
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: address 6 audit findings in auto-loop refactor
1. CRITICAL: Move pendingResolve to AutoSession + queue orphaned agent_end
events instead of silently dropping them. Prevents permanent stalls when
error-recovery sendMessage retries fire between loop iterations.
2. HIGH: Scope pendingResolve per-session via _activeSession ref, preventing
concurrent /gsd auto sessions from corrupting each other's promises.
3. HIGH: Replace console.log in dispatchHookUnit with debugLog to prevent
hook prompt content (potentially containing secrets) from leaking to stdout.
4. HIGH: Restore parked milestone handling in state.ts — Phase 1 skips
parked milestones so they don't satisfy depends_on, Phase 2 registers
them as 'parked' status. Add 'parked' to MilestoneRegistryEntry type.
5. MEDIUM: Restore queuePhaseActive parameter in shouldBlockContextWrite
and re-export setQueuePhaseActive for guided-flow-queue.ts consumers.
6. MEDIUM: Add MAX_LOOP_ITERATIONS (500) lifetime cap to autoLoop to prevent
runaway loops when units alternate between IDs.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: resolve build breakers, add correctness fixes, and graduated recovery
Build breakers (CRITICAL):
- Restore unit-id.ts (deleted but still imported by complexity-classifier.ts, metrics.ts)
- Restore progress-score.ts (deleted but still imported by commands.ts, dashboard-overlay.ts, doctor.ts)
- Rewrite worktree-sync-milestones.test.ts to use new syncProjectRootToWorktree API
Correctness fixes (MEDIUM):
- Cap pendingAgentEndQueue to 3 entries to prevent unbounded growth from stale events
- Add milestoneId path traversal validation in WorktreeResolver
- Clear depthVerificationDone on session_start to prevent cross-session leaks in RPC mode
- Add verification gate for non-hook sidecar units (triage, quick-tasks)
- Remove dead handleAgentEnd import from index.ts
Graduated recovery (Jeremy's feedback):
- Blanket try/catch around loop body — one bad iteration no longer kills the session
- Graduated stuck recovery: at count 3 try artifact verification + cache invalidation,
at count 5 hard stop (was: binary stop at 5 with no recovery attempt)
- Graduated error recovery: 1st error retries, 2nd invalidates caches, 3rd stops
Test results: 32/32 auto-loop, 28/28 worktree-resolver, 11/11 sidecar-queue, tsc clean.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: restore copyWorktreeDb/reconcileWorktreeDb exports and fix loadToolApiKeys import
Two missing exports caused ~90% of the 120 pre-existing test failures:
1. copyWorktreeDb + reconcileWorktreeDb — imported by auto-worktree.ts but
never added to gsd-db.ts. Restored with the original implementations.
2. loadToolApiKeys — moved to commands-config.ts but index.ts still imported
from commands.ts. Fixed the import path.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: move loadToolApiKeys import to commands-config.js
loadToolApiKeys was moved to commands-config.ts but index.ts still
imported it from commands.ts, causing runtime failures in all tests
that transitively load the extension entry point.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* test: fix provider error assertion on windows
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Release session locks on bootstrap abort paths and reset same-process lock state before re-acquiring so stale proper-lockfile callbacks cannot poison a fresh auto-mode session. Adds regression coverage for bootstrap cleanup and re-entrant lock acquisition.
Replace 7 individual ToolResultEvent type guards (isBashToolResult,
isReadToolResult, etc.) with a unified isToolResultEventType() function,
mirroring the existing isToolCallEventType() pattern.
Inline 14 handler type aliases (SendMessageHandler, SetModelHandler, etc.)
directly into the ExtensionActions interface since they were only used there
and added no semantic value.
Update documentation examples to use the new unified guard.
getSessionStats() calculated cost by summing usage from assistant messages
in state.messages. After auto-compaction, pre-compaction messages are
replaced by a compactionSummary with no usage field — dropping the cost.
Fix: Added cumulative accumulators (_cumulativeCost, _cumulativeInputTokens,
_cumulativeOutputTokens, _cumulativeToolCalls) that are incremented on
every assistant message event, independent of the message array.
getSessionStats() now returns max(array-sum, cumulative) to ensure
monotonically non-decreasing values.
Fixes#1423
Three CI regressions from the initial commit:
1. doctor.test.ts "two blocking errors" assertion broke (expected 2, got 3):
The provider check fired on any project with an active milestone, including
CI environments with no API key. Fix: change provider_key_missing severity
from "error" to "warning". A missing key is advisory — it blocks future
dispatch but doesn't corrupt existing state, analogous to env_git_remote.
2. doctor-runtime.test.ts stranded_lock_directory fails on Windows:
proper-lockfile uses advisory file locking on Windows, not the directory-based
mechanism (.gsd.lock/). The check and tests are POSIX-specific. Fix: skip
both stranded_lock_directory tests on Windows with process.platform guard,
same pattern used by worktree and branch tests.
3. doctor-checks.ts used root.split("/").pop() which is not cross-platform:
Windows paths use backslash separators. Fix: replace with basename(root)
from node:path which is platform-aware. Also add basename to imports.
Closes the highest-impact gaps identified in the /gsd doctor deep-dive analysis.
**1. Wire provider checks into runGSDDoctor()**
doctor-providers.ts existed and worked but was never called from the main
doctor run. Units could dispatch into guaranteed API failures with no warning.
Now runProviderChecks() is called in runGSDDoctor() and converts required-provider
errors/warnings into DoctorIssue entries with codes:
- provider_key_missing (error)
- provider_key_backedoff (warning)
**2. Stranded lock directory detection (doctor-checks.ts)**
proper-lockfile creates a .gsd.lock/ directory as the OS-level lock mechanism.
After SIGKILL or hard crash, this directory can remain stranded, blocking all
future auto-mode sessions from acquiring the lock (#1245 pattern). Doctor now:
- Detects .gsd.lock/ existing without a live process holding it
- Reports as stranded_lock_directory (error, fixable)
- Auto-fix removes the stranded directory
**3. Integration branch existence check (doctor-checks.ts + doctor-proactive.ts)**
When a milestone records an integration branch and that branch is later deleted
or renamed, merge-back will fail silently at the end of the milestone. Doctor now:
- Checks each active milestone's stored integration branch exists in git
- Reports as integration_branch_missing (error, not auto-fixable)
- preDispatchHealthGate blocks dispatch if the active milestone's integration
branch is missing, preventing work from being dispatched into a dead end
**4. Orphaned worktree directory detection (doctor-checks.ts)**
Worktree removal can fail after a branch delete, leaving a .gsd/worktrees/<name>/
directory that is no longer registered with git. Re-creating the same name fails
with "already exists". Doctor now:
- Compares .gsd/worktrees/ entries against git worktree list
- Reports unregistered directories as worktree_directory_orphaned (warning, fixable)
- Auto-fix removes the orphaned directory
Tests: all new codes covered with detection + fix assertions, including
false-positive safety cases (live lock holder, registered worktrees,
existing integration branch). All 1843 existing tests still pass.
* fix: sync worktree completion artifacts back to external state before merge (#1412)
When a worktree's .gsd/ was a real directory (not symlinked to external
state), milestone completion artifacts (SUMMARY, VALIDATION, updated
ROADMAP) were written locally but never synced back. The project root's
deriveState() read from external state and found no SUMMARY — reporting
the milestone as incomplete.
Changes:
- auto-worktree.ts: Added syncWorktreeStateBack() that copies milestone
and slice .md files from worktree .gsd/ to the main external state dir
- auto.ts: Call syncWorktreeStateBack() in tryMergeMilestone before the
git merge, ensuring artifacts are visible from the project root
Fixes#1412
* fix: emit agent_end after abort during tool execution (#1414)
When a user aborts a turn while a tool call is running, the abort RPC
succeeds but agent_end was never emitted. RPC consumers tracking turn
lifecycle via events got stuck in a 'streaming' state permanently.
Fix: After abort() + waitForIdle(), emit a synthetic agent_end if the
agent is no longer streaming. This ensures consumers always see the
turn-complete signal regardless of how the turn ended.
Fixes#1414
guided-flow.ts showed 'Interrupted Session Detected' whenever auto.lock
existed, without checking if the lock was written by the current process.
This caused infinite prompt loops when the current session's own lock
triggered the crash detection.
Fix: Added crashLock.pid !== process.pid check, matching the guard in
auto-start.ts.
Also includes test fixes:
- repo-identity-worktree: macOS /var canonicalization
- resource-loader: partial-build dist/resources fallback
- file-watcher: init delay + timeout for timing stability
Fixes#1398
Two root causes for the false "Interrupted Session Detected" prompt
that appears every time /gsd is run after a normal exit:
1. guided-flow.ts showed the crash recovery menu even for bootstrap
crashes (unitType="starting", unitId="bootstrap", completedUnits=0)
where no work was lost. Now these are silently discarded — the menu
only appears when real auto-mode work was interrupted.
2. session-lock.ts exit handler cleaned the OS lock directory
(.gsd.lock/) but not the auto.lock metadata file. On next startup,
readCrashLock() found the stale file and triggered false recovery.
Now the exit handler also removes auto.lock.
