The previous guard `if (after.trim().length > 0)` incorrectly triggered
when `after` contained only footer text (e.g. `---\n*Last updated*`).
In that case `after.replace(pattern, replacement)` is a no-op and the
function returned unchanged content instead of falling through to the
slow path that searches inside the last `<details>` block.
Fix: capture the replaced string first, then only take the fast path
when the replacement actually changed `after`.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds a case where the active milestone is the last <details> block but
footer text (--- / *Last updated*) follows </details>, triggering the
fast-path to replace in the footer instead of inside the block.
Closes#2743
- Add _deepMergeConfig() with correct null-override semantics
- loadConfig() reads root config.json when GSD_WORKSTREAM is set, then
deep-merges with workstream config (workstream wins on conflict)
- Workstream without config.json falls back to root config entirely
- Migrations and disk writes operate on fileData (on-disk content) only,
never on the merged result, to prevent workstream pollution
- Fixes null-override bug from PR #2717: explicit null in workstream now
correctly overrides root value instead of falling back to root
- Tests: inherit root model_overrides, workstream override, nested
workflow.* deep merge, explicit null override, missing workstream config
Closes#2714
replaceInCurrentMilestone's lastIndexOf('</details>') heuristic fails
when the active milestone itself is wrapped in a <details> block — the
after-slice is empty so the replacement is silently dropped.
Fix detects this case (after.trim().length === 0) and falls back to
locating the last complete <details>…</details> span and applying the
replacement only inside it, leaving all earlier archived-milestone
blocks untouched.
Closes#2641
The SDK's buildExecutorPrompt told executors to "Create a SUMMARY.md file"
with no directory path, causing them to write it in cwd (project root)
instead of .planning/phases/{phase}/. Thread phaseDir from PhaseRunner
through PromptFactory and into the completion instructions so the executor
gets an explicit path like `.planning/phases/01-auth/01-01-SUMMARY.md`.
Backward compatible — buildExecutorPrompt still accepts a plain string
(agentDef) for existing callers.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(#2722): forensics gh commands pin --repo gsd-build/get-shit-done
gh issue create and gh label list both defaulted to the repo inferred
from $PWD, causing issues to be submitted to the user's current project
instead of this repo. Added --repo gsd-build/get-shit-done to both
commands. Added two regression tests covering both gh calls.
Closes#2722
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(#2723): scope forensics tests to specific gh commands, not whole file
CodeRabbit found that the gh issue create test searched the whole
workflow file, so it would pass even if gh issue create lacked --repo
(because gh label list already contains the repo string elsewhere).
Also replaced the brittle 200-char slice in the label-list test with
a regex. Both tests now use assert.match() with command-scoped regexes.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
377a6d2 deleted sdk/prompts/agents/ and sdk/prompts/workflows/ (13 files)
but did not update 3 test files that reference them, causing ENOENT
failures on every CI run (main and all PRs) since that commit.
Removed:
- sdk/prompts/agents variants describe block (enh-2427-sycophancy-hardening)
- PLAN_PHASE_SDK_PATH constant and headless plan-phase test (post-planning-gaps-2493)
- sdk/prompts/workflows/verify-phase.md describe block (verifier-deferred-items)
The underlying behaviour is covered by the existing main agent/workflow
tests; the SDK variant tests are moot now that the SDK loads installed
files instead of bundled stripped-down copies.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
The discuss step loaded the full interactive workflow prompt which
instructs the agent to use AskUserQuestion, Skill(), and area selection
UIs. In headless auto mode, the agent followed these instructions and
tried to interact with a non-existent user.
Fix: prepend a mandatory headless override BEFORE the workflow prompt
that explicitly forbids interactive tools and instructs the agent to
make all decisions autonomously. Prepending (not appending) ensures
the override takes priority over conflicting instructions later in
the prompt.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Bug 1: phasePlanIndex derived empty planId for bare PLAN.md files.
Fixed to use 'PLAN' as the ID, with matching SUMMARY.md detection.
Bug 2: executeSinglePlan passed null to buildPrompt instead of the
actual parsed plan. The executor needs the plan content (tasks,
objectives) to know what to build. Now loads and parses the plan
file before building the prompt.
Bug 3: parseVerificationOutcome checked session exit code, not what
the verifier wrote. A session that runs without errors but writes
status: gaps_found to VERIFICATION.md was treated as 'passed'. Now
queries check.verification-status to read the actual VERIFICATION.md
frontmatter status field.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The SDK bundled its own agents and workflows at ~17% the size of the real
ones, missing critical instructions like file naming conventions, scope
reduction rules, discovery protocols, and TDD integration. This caused
the planner to create a single PLAN.md instead of properly named
per-plan files (01-01-PLAN.md, 01-02-PLAN.md), breaking wave-based
parallel execution.
- Invert load priority: installed GSD agents/workflows first, SDK
bundled as last-resort fallback
- Replace @-reference stripping with resolution (read + inline content)
- Use full agent definitions instead of extracting only the <role> block
- Delete sdk/prompts/agents/ and sdk/prompts/workflows/ (13 files)
- Delete headless-prompts.test.ts (validated deleted files)
- Thread projectDir through sanitizePrompt for @-reference resolution
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* feat(#2500): enrich gsd-codebase-mapper arch-focus ARCHITECTURE.md template
The codebase mapper's arch-focus template was a sparse structural inventory.
After major refactors, the research/ARCHITECTURE.md (created at /gsd-new-project
and never refreshable) went stale while the refreshable codebase version lacked
the visual richness that makes architecture docs useful for planning.
Add to the ARCHITECTURE.md template:
- <!-- refreshed: {date} --> marker at the top (maintainer request)
- ASCII system overview diagram with component boxes and flow arrows
- Component responsibility table (Component / Responsibility / File)
- Primary request path traces with numbered steps and code references
- Architectural constraints section (threading, global state, circular imports)
- Anti-patterns section with codebase-specific patterns and correct alternatives
All existing sections (Pattern Overview, Layers, Key Abstractions, Entry Points,
Error Handling, Cross-Cutting Concerns) are preserved.
7 new tests in tests/enh-2500-codebase-mapper-arch-rich-format.test.cjs verify
each required section is present in the deployed template.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(#2500): resolve CodeRabbit review findings
- Add 'text' language tag to bare ASCII diagram fenced block (markdownlint MD040)
- Tighten data flow test: require '### Primary Request Path' heading, 3+
numbered steps, and file:line reference pattern — prevents loose-match
false positives
- Tighten constraints test: require '## Architectural Constraints' heading
AND Threading / Global state / Circular imports tokens — prevents broad
keyword matches masking regressions
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* feat(#2306): plan-review-convergence v2 — CYCLE_SUMMARY contract, config gate, local model reviewers
Fixes the false-stall detection bug in the plan→review→replan convergence
loop. REVIEWS.md accumulates history across cycles so raw grep inflated
HIGH counts; HIGH count now comes from a per-cycle CYCLE_SUMMARY contract
emitted in the review agent's return message.
Key changes:
- workflow.plan_review_convergence config gate (disabled by default, same
pattern as workflow.code_review / workflow.nyquist_validation)
- Review agent prompt defines CYCLE_SUMMARY: current_high=<N> contract with
PARTIALLY RESOLVED / FULLY RESOLVED counting rules
- Orchestrator aborts on absent/malformed CYCLE_SUMMARY (distinguishes both)
- Warns when HIGH_COUNT > 0 but ## Current HIGH Concerns section is missing
- Stall detection and --ws forwarding preserved and tested
- Local model reviewers: --ollama, --lm-studio, --llama-cpp flags added to
convergence workflow and review workflow; all three use OpenAI-compatible
/v1/chat/completions endpoint with jq --rawfile for safe JSON encoding
- review.ollama_host / review.lm_studio_host / review.llama_cpp_host config
keys registered and documented (default to localhost:11434/1234/8080)
- review.models.ollama / .lm_studio / .llama_cpp model-name config support
- 58 tests (up from 29 in PR #2339), all passing
Closes#2306Closes#2339
Co-authored-by: Tom Boucher <trekkie@nomorestars.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(ci): sync sdk/src/query/config-schema.ts with CJS schema (#2306)
Add workflow.plan_review_convergence, review.ollama_host,
review.lm_studio_host, and review.llama_cpp_host to the SDK-side
TypeScript mirror — required by the CJS↔SDK parity test (#2653).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(#2306): resolve CodeRabbit review findings
- Anchor HIGH_COUNT extraction with head -1 to prevent multi-match when
agent return message contains multiple CYCLE_SUMMARY lines (e.g. quoted
back from prompt context)
- Replace hardcoded reviewers list in REVIEWS.md frontmatter template with
runtime-derived placeholder — the static list did not reflect which
reviewers were actually invoked
- Broaden workflow.plan_review_convergence docs to include local reviewers
(Ollama, LM Studio, llama.cpp) alongside cloud reviewers
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(ci): restore reviewers frontmatter list with runtime note
The cursor-reviewer.test.cjs (and equivalent per-reviewer tests) assert
that each supported reviewer appears on the reviewers: line — these are
wiring tests that catch when a new reviewer is added to invocation but
not to the REVIEWS.md template. Replacing the list with a placeholder
broke those tests.
Restore the full static list and add an inline comment clarifying that
the actual committed frontmatter should be filtered to only the reviewers
invoked that run — satisfying both the per-reviewer tests and the
CodeRabbit correctness note.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Fixes#2698 — The two separate LF/CRLF .replace() calls in the Codex hooks
migration could not handle mixed line endings (e.g. header in LF, body in
CRLF), leaving stale gsd-update-check blocks after reinstall. Consolidated to
a single \r?\n-aware regex with gm flags that handles LF, CRLF, and mixed
content in one pass.
Fixes#2678 — installSdkIfNeeded() called process.exit(1) unconditionally when
sdk/dist/cli.js was missing, even during --local installs where users cannot
write to global node_modules. Added isLocal option: when true, prints a warning
and returns instead of exiting.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(#2686): review-fix agent now uses git worktree for isolation
The gsd-code-fixer agent operated directly against the main working tree,
racing any concurrent foreground session for HEAD, the index, and on-disk
files. Added a setup_worktree step (git worktree add /tmp/sv-N-reviewfix
HEAD) as the first action before any file operations, with unconditional
git worktree remove cleanup on exit. Mirrors the pattern used by all other
GSD per-issue agents.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(#2686): address CodeRabbit review — mktemp unique path, branch-aware worktree, tighten test assertions
- Use mktemp -d for unique worktree path (prevents concurrent-run collision)
- Resolve branch via git branch --show-current before worktree add (prevents detached HEAD)
- Error-and-exit on worktree add failure instead of force-removing shared path
- Test: use .exec().index for checkout position (not indexOf on match string)
- Test: match gsd-sdk query commit as well as git commit for ordering assertion
- Test: tighten /tmp path assertion to require actual /tmp/sv- assignment
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* test(#2692): add behavioral --wave N test, annotate source-text assertions
Adds two behavioral tests for wave filtering via phase-plan-index:
- Verifies plans with wave frontmatter are correctly grouped by wave number
- Verifies plans with no wave field default to wave 1
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(#2684,#2676): milestone.complete version validation + parallel milestone phase routing
#2684: Confirms milestone.complete correctly validates and uses its version
argument end-to-end. The inline archive path in milestoneComplete already
forwarded version correctly; regression tests lock in that contract.
#2676: phase.complete applied getMilestonePhaseFilter unconditionally, using
STATE.md's primary milestone to scope the candidate set. When the completed
phase belongs to a parallel (secondary) milestone, the filter excluded all
phases from that milestone, leaving an empty candidate set and incorrectly
returning is_last_phase: true / next_phase: null.
Fix: before applying the milestone filter in Step E, check whether the
completed phase itself appears in the filtered set. If not, skip the filter
for both the directory scan and the ROADMAP.md fallback so phases from the
secondary milestone remain visible for next-phase detection.
Closes#2684Closes#2676
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
VALID_PROFILES was derived solely from Object.keys(MODEL_PROFILES['gsd-planner']),
which only contained the named tiers (quality/balanced/budget/adaptive). The
cmdConfigSetModelProfile validator rejected 'inherit' even though the runtime has
supported it since #1829. Fix: append 'inherit' to VALID_PROFILES and handle it
in getAgentToModelMapForProfile so the agent→model table shows 'inherit' instead
of undefined.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds two behavioral tests for wave filtering via phase-plan-index:
- Verifies plans with wave frontmatter are correctly grouped by wave number
- Verifies plans with no wave field default to wave 1
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* docs(test-standards): enforce no-source-grep rule with CI linter + update CONTRIBUTING.md
Adds scripts/lint-no-source-grep.cjs — a static linter that detects readFileSync
on .cjs source files in tests without an allow-test-rule annotation. Wires it
into CI as a new lint-tests job in test.yml and as npm run lint:tests.
Resolves all 9 existing violations across the test suite:
- Rewrites workspace routing tests (3) as behavioral runGsdTools calls that
verify each command is router-recognized (exit != "Unknown init workflow")
- Adds allow-test-rule annotations with explanatory comments to 7 legitimate
structural tests: architectural invariants (locking, orphan-worktree),
structural regression guards (milestone-regex-global), docs-parity
(config-field-docs), integration-test-input (copilot-install), and
structural-implementation-guards (bug-1891, discuss-mode)
Updates CONTRIBUTING.md Testing Standards section with:
- "Prohibited: Source-Grep Tests" section with the before/after pattern,
root cause analysis of why it breaks (commit 990c3e64), and CI reference
- allow-test-rule exemption table (6 recognized categories with when-to-use)
- "CI Test Quality Checks" table showing lint-tests job and local run command
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix: resolve CodeRabbit findings on PR #2700
- CONTRIBUTING.md: "four recognized categories" → "six" (table has 6 rows)
- workspace.test.cjs: use positional args in routing tests (no --name flag)
- lint-no-source-grep.cjs: add source-dir guard to READ_WITH_INLINE_CJS_RE
(mirrors CJS_PATH_CONST_RE's protection against false positives on temp files)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(lint): tighten allow-test-rule and add recursive test discovery
- ALLOW_ANNOTATION now requires at least one non-whitespace char after the
colon so bare '// allow-test-rule:' cannot bypass the lint gate
- findTestFiles() recurses into subdirectories so nested *.test.cjs files
are covered if the tests/ tree ever grows subdirs
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
All workflow, command, reference, template, and tool-output files that
surfaced /gsd:<cmd> as a user-typed slash command have been updated to
use /gsd-<cmd>, matching the Claude Code skill directory name.
Closes#2697
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* test: destroy 9 config-schema.cjs/core.cjs source-grep tests, add behavioral config-set tests (#2691, #2693)
Replace source-grep theater with config-set behavioral tests:
- execute-phase-wave: config-set workflow.use_worktrees replaces VALID_CONFIG_KEYS grep
- inline-plan-threshold: delete redundant source-grep (behavioral test at L36 already covered it)
- plan-bounce: config-set for plan_bounce / plan_bounce_script / plan_bounce_passes replaces 3 key-presence greps
- code-review: config-set for code_review / code_review_depth replaces 2 greps; removes CONFIG_PATH constant
- thinking-partner: config-set features.thinking_partner replaces two greps (config-schema.cjs AND core.cjs)
Behavioral tests survive refactors (no path constants, no file reads). The config-schema.cjs →
core.cjs migration commit 990c3e64 happened because these tests groped source paths.
Add allow-test-rule: source-text-is-the-product annotations to legitimate product-content tests:
autonomous-allowed-tools, agent-frontmatter, agent-skills-awareness, bug-2334, bug-2346,
execute-phase-wave (MD reads), plan-bounce (workflow reads). Annotations explain WHY text
inspection is the right level of testing for AI instruction files.
Closes#2691Closes#2693
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test: address CodeRabbit findings on #2696
- agent-frontmatter.test.cjs: move allow-test-rule annotation from block comment
to standalone // line comment so rule scanners can detect it
- thinking-partner.test.cjs: strengthen config-set test with config-get read-back
assertion to verify the value was persisted, not just accepted (exit 0)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test: tighten thinking_partner config assertion per CodeRabbit (#2696)
Replace config-get output substring check (includes('true') false-positive
risk) with a direct JSON read of .planning/config.json, asserting the
exact persisted value via strictEqual. This also validates the config file
was created, catching silent key-acceptance without persistence.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(#2661): unconditional plan-checkbox sync in execute-plan
Checkpoint A in execute-plan.md was wrapped in a "Skip in parallel mode"
guard that also short-circuited the parallelization-without-worktrees
case. With `parallelization: true, use_worktrees: false`, only
Checkpoint C (phase.complete) then remained, and any interruption
between the final SUMMARY write and phase complete left ROADMAP.md
plan checkboxes stale.
Remove the guard: `roadmap update-plan-progress` is idempotent and
atomically serialized via readModifyWriteRoadmapMd's lockfile, so
concurrent invocations from parallel plans converge safely.
Checkpoint B (worktree-merge post-step) and Checkpoint C
(phase.complete) become redundant after A is unconditional; their
removal is deferred to a follow-up per the RCA.
Closes#2661
* fix(#2661): gate ROADMAP sync on use_worktrees=false to preserve single-writer contract
Adversarial review of PR #2682 found that unconditionally removing the
IS_WORKTREE guard violates the single-writer contract for shared
ROADMAP.md established by commit dcb50396 (PR #1486). The lockfile only
serializes within a single working tree; separate worktrees have
separate ROADMAP.md files that diverge.
Restore the worktree guard but document its intent explicitly: the
in-handler sync runs only when use_worktrees=false (the actual #2661
reproducer). Worktree mode relies on the orchestrator's post-merge
update at execute-phase.md lines 815-834, which is the documented
single-writer for shared tracking files.
Update tests to assert both branches of the gate:
- use_worktrees: false mode runs the sync (the #2661 case)
- use_worktrees: true mode does NOT run the in-handler sync
- handler-level idempotence and lockfile contention tests retained,
scope clarified to within-tree concurrency only
* fix(#2660): capture prose after label in extractOneLinerFromBody
The regex `\*\*([^*]+)\*\*` matched the first bold span, so for the new
SUMMARY template `**One-liner:** Real prose here.` it captured the label
`One-liner:` instead of the prose. MILESTONES.md then wrote bullets like
`- One-liner:` with no content.
Handle both template forms:
- Labeled: `**One-liner:** prose` → prose
- Bare: `**prose**` → prose (legacy)
Empty prose after a label returns null so no bogus bullets are emitted.
Note: existing MILESTONES.md entries generated under the bug are not
regenerated here — that is a follow-up.
Closes#2660
* fix(#2660): normalize CRLF before one-liner extraction
Windows-authored SUMMARY files use CRLF line endings; the LF-only regex
in extractOneLinerFromBody would fail to match. Normalize \r\n and \r
to \n before stripping frontmatter and matching the one-liner pattern.
Adds test case (h) covering CRLF input.
* fix(#2659): qualify bare output() calls in audit-open handler
The audit-open dispatch case in bin/gsd-tools.cjs previously called bare
output() on both --json and text branches, which crashed with
ReferenceError: output is not defined. The core module is imported as
`const core`, so every other case uses core.output(). HEAD already
qualifies the calls correctly; this commit adds a regression test that
invokes `audit-open` and `audit-open --json` through runGsdTools and
asserts a clean exit plus non-empty stdout (and an explicit check that
the failure mode is not ReferenceError). The test fails on any revision
where either call reverts to bare output().
Closes#2659
* test(#2659): assert valid JSON output in --json mode
CodeRabbit nit: tighten --json regression coverage by parsing stdout
and asserting the result is a JSON object/array, not just non-empty.
initProgress computed phase status purely from disk (PLAN/SUMMARY counts),
consulting the ROADMAP `- [x] Phase N` checkbox only for phases with no
directory. initManager, by contrast, applied an explicit override: a
ROADMAP `[x]` forces status to `complete` regardless of disk state.
Result: a phase with a stub directory (no SUMMARY.md) and a ticked
ROADMAP checkbox reported `complete` from /gsd-manager and `pending`
from /gsd-progress — same data, different answer.
Apply ROADMAP-[x]-wins as the unified policy inside initProgress, mirroring
initManager's override. A user who typed `- [x] Phase 3` has made an
explicit assertion; a leftover stub dir is the weaker signal.
Adds sdk/src/query/init-progress-precedence.test.ts covering six cases
(stub dir + [x], full dir + [x], full dir + [ ], stub dir + [ ],
ROADMAP-only + [x], and completed_count parity). Pre-fix: cases 1 and 6
failed. Post-fix: all six pass. No existing tests were modified.
Closes#2674
/gsd-insert-phase step 4 instructed the agent to directly Edit/Write
.planning/STATE.md to append a Roadmap Evolution entry. Projects that
ship a protect-files.sh PreToolUse hook (a recommended hardening
pattern) blocked the raw write, silently leaving STATE.md out of sync
with ROADMAP.md.
Adds a dedicated SDK handler state.add-roadmap-evolution (plus space
alias) that:
- Reads STATE.md through the shared readModifyWriteStateMd lockfile
path (matches sibling mutation handlers — atomic against
concurrent writers).
- Locates ### Roadmap Evolution under ## Accumulated Context, or
creates both sections as needed.
- Dedupes on exact-line match so idempotent retries are no-ops
({ added: false, reason: "duplicate" }).
- Validates --phase / --action presence and action membership,
throwing GSDError(Validation) for bad input (no silent
{ ok: false } swallow).
Workflow change (insert-phase.md step 4):
- Replaces the raw Edit/Write instructions for STATE.md with
gsd-sdk query state.patch (for the next-phase pointer) and
gsd-sdk query state.add-roadmap-evolution (for the evolution
log).
- Updates success criteria to check handler responses.
- Drops "Write" from commands/gsd/insert-phase.md allowed-tools
(no step in the workflow needs it any more).
Tests (vitest, sdk/src/query/state-mutation.test.ts): subsection
creation when missing; append-preserving-order when present;
duplicate -> reason=duplicate; idempotence over two calls; three
validation cases covering missing --phase, missing --action, and
invalid action.
This is the first SDK handler dedicated to STATE.md Roadmap
Evolution mutations. Other workflows with similar raw STATE.md
edits (/gsd-pause-work, /gsd-resume-work, /gsd-new-project,
/gsd-complete-milestone, /gsd-add-phase) remain on raw Edit/Write
and will need follow-up issues to migrate — out of scope for this
fix.
Closes#2662
* fix(#2633): use ROADMAP.md as authority for current-milestone phase counts
initMilestoneOp (SDK + CJS) derives phase_count and completed_phases from
the current milestone section of ROADMAP.md instead of counting on-disk
`.planning/phases/` directories. After `phases clear` at the start of a new
milestone the on-disk set is a subset of the roadmap, causing premature
`all_phases_complete: true`.
validateHealth W002 now unions ROADMAP.md phase declarations (all milestones
— current, shipped, backlog) with on-disk dirs when checking STATE.md phase
refs. Eliminates false positives for future-phase refs in the current
milestone and history-phase refs from shipped milestones.
Falls back to legacy on-disk counting when ROADMAP.md is missing or
unparseable so no-roadmap fixtures still work.
Adds vitest regressions for both handlers; all 66 SDK + 118 CJS tests pass.
* fix(#2633): preserve full phase tokens in W002 + completion lookup
CodeRabbit flagged that the parseInt-based normalization collapses distinct
phase IDs (3, 3A, 3.1) into the same integer bucket, masking real
STATE/ROADMAP mismatches and miscounting completions in milestones with
inserted/sub-phases.
Index disk dirs and validate STATE.md refs by canonical full phase token —
strip leading zeros from the integer head only, preserve [A-Z] suffix and
dotted segments, and accept just the leading-zero variant of the integer
prefix as a tolerated alias. 3A and 3 never share a bucket.
Also widens the disk and STATE.md regexes to accept [A-Z]? suffix tokens.
* fix(#2636): surface gsd-sdk query failures and add workflow↔handler parity check
Root cause: workflows invoked `gsd-sdk query agent-skills <slug>` with a
trailing `2>/dev/null`, swallowing stderr and exit code. When the installed
`@gsd-build/sdk` npm was stale (pre-query), the call resolved to an empty
string and `agent_skills.<slug>` config was never injected into spawn
prompts — silently. The handler exists on main (sdk/src/query/skills.ts),
so this is a publish-drift + silent-fallback bug, not a missing handler.
Fix:
- Remove bare `2>/dev/null` from every `gsd-sdk query agent-skills …`
invocation in workflows so SDK failures surface to stderr.
- Apply the same rule to other no-fallback calls (audit-open, write-profile,
generate-* profile handlers, frontmatter.get in commands). Best-effort
cleanup calls (config-set workflow._auto_chain_active false) keep
exit-code forgiveness via `|| true` but no longer suppress stderr.
Parity tests:
- New: tests/bug-2636-gsd-sdk-query-silent-swallow.test.cjs — fails if any
`gsd-sdk query agent-skills … 2>/dev/null` is reintroduced.
- Existing: tests/gsd-sdk-query-registry-integration.test.cjs already
asserts every workflow noun resolves to a registered handler; confirmed
passing post-change.
Note: npm republish of @gsd-build/sdk is a separate release concern and is
not included in this PR.
* fix(#2636): address review — restore broken markdown fences and shell syntax
The previous commit's mass removal of '2>/dev/null' suffixes also
collapsed adjacent closing code fences and 'fi' tokens onto the
command line, producing malformed markdown blocks and 'truefi' /
'true fi' shell syntax errors in the workflows.
Repaired sites:
- commands/gsd/quick.md, thread.md (frontmatter.get fences)
- workflows/complete-milestone.md (audit-open fence)
- workflows/profile-user.md (write-profile + generate-* fences)
- workflows/verify-work.md (audit-open --json fence)
- workflows/execute-phase.md (truefi -> true / fi)
- workflows/plan-phase.md, discuss-phase-assumptions.md,
discuss-phase/modes/chain.md (true fi -> true / fi)
All 5450 tests pass.
* fix(#2645): emit [[agents]] array-of-tables in Codex config.toml
Codex ≥0.116 rejects `[agents.<name>]` map tables with `invalid type:
map, expected a sequence`. Switch generateCodexConfigBlock to emit
`[[agents]]` array-of-tables with an explicit `name` field per entry.
Strip + merge paths now self-heal on reinstall — both the legacy
`[agents.gsd-*]` map shape (pre-#2645 configs) and the new
`[[agents]]` with `name = "gsd-*"` shape are recognized and replaced,
while user-authored `[[agents]]` entries are preserved.
Fixes#2645
* fix(#2645): use TOML-aware parser to strip managed [[agents]] sections
CodeRabbit flagged that the prior regex-based stripper for [[agents]]
array-of-tables only matched headers at column 0 and stopped at any line
beginning with `[`. An indented [[agents]] header would not terminate the
preceding match, so a managed `gsd-*` block could absorb a following
user-authored agent and silently delete it.
Replace the ad-hoc regex with the existing TOML-aware section parser
(getTomlTableSections + removeContentRanges) so section boundaries are
authoritative regardless of indentation. Same logic applies to legacy
[agents.gsd-*] map sections.
Add a comprehensive mixed-shape test covering multiple GSD entries (both
legacy map and new array-of-tables, double- and single-quoted names)
interleaved with multiple user-authored agents in both shapes — verifies
all GSD entries are stripped and every user entry is preserved.
* fix(#2652): layer ~/.gsd/defaults.json over built-ins in SDK loadConfig
SDK loadConfig only merged built-in CONFIG_DEFAULTS, so pre-project init
queries (e.g. resolveModel in Codex installs) ignored user-level knobs like
resolve_model_ids: "omit" and emitted Claude model aliases from MODEL_PROFILES.
Port the user-defaults layer from get-shit-done/bin/lib/config.cjs:65 to the
TS loader. CJS parity: user defaults only apply when no .planning/config.json
exists (buildNewProjectConfig already bakes them in at /gsd:new-project time).
Fixes#2652
* fix(#2652): isolate GSD_HOME in test, refresh loadConfig JSDoc (CodeRabbit)
installCodexConfig() applied a narrow path-only regex pass before
generateCodexAgentToml(), skipping the convertClaudeToCodexMarkdown() +
neutralizeAgentReferences(..., 'AGENTS.md') pipeline used on the .md emit
path. Result: emitted Codex agent TOMLs carried stale Claude-specific
references (CLAUDE.md, .claude/skills/, .claude/commands/, .claude/agents/,
.claudeignore, bare "Claude" agent-name mentions).
Route the TOML path through convertClaudeToCodexMarkdown and extend that
pipeline to cover bare .claude/<subdir>/ references and .claudeignore
(both previously unhandled on the .md path too). The $HOME/.claude/
get-shit-done prefix substitution still runs first so the absolute Codex
install path is preserved before the generic .claude → .codex rewrite.
Regression test: tests/issue-2639-codex-toml-neutralization.test.cjs —
drives installCodexConfig against a fixture containing every flagged
marker and asserts the emitted TOML contains zero CLAUDE.md / .claude/
/ .claudeignore occurrences and that Claude Code / Claude Opus product
names survive.
Fixes#2639
initProgress (and its CJS twin) hardcoded `not_started` for ROADMAP-only
phases, so `completed_count` stayed at 0 even when the ROADMAP showed
`- [x] Phase N`. Extract ROADMAP checkbox states into a shared helper
and use `- [x]` as the completion signal when no phase directory is
present. Disk status continues to win when both exist.
Adds a regression test that reproduces the bug with no phases/ dir and
one `[x]` / one `[ ]` phase, asserting completed_count===1.
Fixes#2646
Flat-skills installs write SKILL.md files under gsd-<cmd>/ dirs, but
Claude Code resolves skills by their frontmatter `name:`, not directory
name. PR #2595 normalized every `/gsd-<cmd>` to `/gsd:<cmd>` across
workflows — including inside `Skill(skill="...")` args — but the
installer still emitted `name: gsd-<cmd>`, so every Skill() call on a
flat-skills install resolved to nothing.
Fix: emit `name: gsd:<cmd>` (colon form) in
`convertClaudeCommandToClaudeSkill`. Keep the hyphen-form directory
name for Windows path safety.
Codex stays on hyphen form: its adapter invokes skills as `$gsd-<cmd>`
(shell-var syntax) and a colon would terminate the variable name.
`convertClaudeCommandToCodexSkill` uses `yamlQuote(skillName)` directly
and is untouched.
- Extract `skillFrontmatterName(dirName)` helper (exported for tests).
- Update claude-skills-migration and qwen-skills-migration assertions
that encoded the old hyphen emission.
- Add `tests/bug-2643-skill-frontmatter-name.test.cjs` asserting every
`Skill(skill="gsd:<cmd>")` reference in workflows resolves to an
emitted frontmatter name.
Full suite: 5452/5452 passing.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
loadConfig's multiRepo migration and filesystem-sync writers targeted the
top-level parsed.sub_repos, but KNOWN_TOP_LEVEL (the unknown-key validator's
allowlist) only recognizes planning.sub_repos (canonical per #2561). Each
migration/sync therefore persisted a key the next loadConfig call warned was
unknown.
Redirect both writers to parsed.planning.sub_repos, ensuring parsed.planning
is initialized first. Also self-heal legacy/buggy installs by stripping any
stale top-level sub_repos on load, preserving its value as the
planning.sub_repos seed if that slot is empty.
Tests cover: (a) canonical planning.sub_repos emits no warning, (b) multiRepo
migration writes to planning.sub_repos with no top-level residue,
(c) filesystem sync relocates to planning.sub_repos, (d) stale top-level
sub_repos from older buggy installs is stripped on load.
Closes#2638
v1.38.3 shipped without sdk/dist/ because the outer `files` whitelist
and `prepublishOnly` chain had drifted. The `gsd-sdk` bin shim then
fell through to a stale @gsd-build/sdk@0.1.0 (pre-`query`), breaking
every workflow that called `gsd-sdk query <noun>` on fresh installs.
Current package.json already restores `sdk/dist` + `build:sdk`
prepublish; this PR locks the fix in with:
- tests/bug-2647-outer-tarball-sdk-dist.test.cjs — asserts `files`
includes `sdk/dist`, `prepublishOnly` invokes `build:sdk`, the
shim resolves sdk/dist/cli.js, `npm pack --dry-run` lists
sdk/dist/cli.js, and the built CLI exposes a `query` subcommand.
- scripts/verify-tarball-sdk-dist.sh — packs, extracts, installs
prod deps, and runs `node sdk/dist/cli.js query --help` against
the real tarball output.
- .github/workflows/release.yml — runs the verify script in both
next and stable release jobs before `npm publish`.
Partial fix for #2649 (same root cause on the sibling sdk package).
Fixes#2647
The SDK's config-set kept its own hand-maintained allowlist (28-key
drift vs. get-shit-done/bin/lib/config-schema.cjs), so documented
keys accepted by the CJS config-set — planning.sub_repos,
workflow.code_review_command, workflow.security_*, review.models.*,
model_profile_overrides.*, etc. — were rejected with
"Unknown config key" when routed through the SDK.
Changes:
- New sdk/src/query/config-schema.ts mirrors the CJS schema exactly
(exact-match keys + dynamic regex sources).
- config-mutation.ts imports VALID_CONFIG_KEYS / DYNAMIC_KEY_PATTERNS
from the shared module instead of rolling its own set and regex
branches.
- Drop hand-coded agent_skills.* / features.* regex branches —
now schema-driven so claude_md_assembly.blocks.*, review.models.*,
and model_profile_overrides.<runtime>.<tier> are also accepted.
- Add tests/config-schema-sdk-parity.test.cjs (node:test) as the
CI drift guard: asserts CJS VALID_CONFIG_KEYS set-equals the
literal set parsed from config-schema.ts, and that every CJS
dynamic pattern source has an identical counterpart in the SDK.
Parallel to the CJS↔docs parity added in #2479.
- Vitest #2653 specs iterate every CJS key through the SDK
validator, spot-check each dynamic pattern, and lock in
planning.sub_repos.
- While here: add workflow.context_coverage_gate to the CJS schema
(already in docs and SDK; CJS previously rejected it) and sync
the missing curated typo-suggestions (review.model, sub_repos,
plan_checker, workflow.review_command) into the SDK.
Fixes#2653.
The /gsd:new-milestone workflow Step 5 rewrote STATE.md's Current Position
body but never touched the YAML frontmatter, so every downstream reader
(state.json, getMilestoneInfo, progress bars) kept reporting the stale
milestone until the first phase advance forced a resync. Asymmetric with
milestone.complete, which uses readModifyWriteStateMdFull.
Add a new `state milestone-switch` handler (both SDK and CJS) that atomically:
- Stomps frontmatter milestone/milestone_name with caller-supplied values
- Resets status to 'planning' and progress counters to zero
- Rewrites the ## Current Position section to the new-milestone template
- Preserves Accumulated Context (decisions, blockers, todos)
Wire the workflow Step 5 to invoke `state.milestone-switch` instead of the
manual body rewrite. Note the flag is `--milestone` not `--version`:
gsd-tools reserves `--version` as a globally-invalid help flag.
Red vitest in sdk/src/query/state-mutation.test.ts asserts the frontmatter
reset. Regression guard via node:test in tests/bug-2630-*.test.cjs runs
through gsd-tools end-to-end.
Fixes#2630
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Root cause shared with #2647: a broken 1.38.3 tarball shipped without
sdk/dist/. The pre-#2441-decouple installer reacted by running
spawnSync('npm.cmd', ['install'], { cwd: sdkDir }) inside the npx cache
on Windows, where the cache is read-only, producing the misleading
"Failed to npm install in sdk/" error.
Defensive changes here (user-facing behavior only; packaging fix lives
in the sibling PR for #2647):
- Classify the install context (classifySdkInstall): detect npx cache
paths, node_modules-based installs, and dev clones via path heuristics
plus a side-effect-free write probe. Exported for test.
- Rewrite the dist-missing error to branch on context:
tarball + npxCache -> "don't touch npx cache; npm i -g ...@latest"
tarball (other) -> upgrade path + clone-build escape hatch
dev-clone -> keep existing cd sdk && npm install && npm run build
- Preserve the invariant that the installer never shells out to
npm install itself — users always drive that.
- Add tests/bug-2649-sdk-fail-fast.test.cjs covering the classifier and
both failure messages, with spawnSync/execSync interceptors that
assert no nested npm install is attempted.
Cross-ref: #2647 (packaging).
Fixes#2649
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(workflows): agent-skills query keys must match subagent_type
Eight workflow files called `gsd-sdk query agent-skills <KEY>` with
a key that did not match any `subagent_type` Task() spawns in the
same workflow (or any existing `agents/<KEY>.md`):
- research-phase.md:45 — gsd-researcher → gsd-phase-researcher
- plan-phase.md:36 — gsd-researcher → gsd-phase-researcher
- plan-phase.md:38 — gsd-checker → gsd-plan-checker
- quick.md:145 — gsd-checker → gsd-plan-checker
- verify-work.md:36 — gsd-checker → gsd-plan-checker
- new-milestone.md:207 — gsd-synthesizer → gsd-research-synthesizer
- new-project.md:63 — gsd-synthesizer → gsd-research-synthesizer
- ui-review.md:21 — gsd-ui-reviewer → gsd-ui-auditor
- discuss-phase.md:114 — gsd-advisor → gsd-advisor-researcher
Effect before this fix: users configuring `agent_skills.<correct-type>`
in .planning/config.json got no injection on these paths because the
workflow asked the SDK for a different (non-existent) key. The SDK
correctly returned "" for the unknown key, which then interpolated as
an empty string into the Task() prompt. Silent no-op.
The discuss-phase advisor case is a subtle variant — the spawn site
uses `subagent_type="general-purpose"` and loads the agent role via
`Read(~/.claude/agents/gsd-advisor-researcher.md)`. The injection key
must follow the agent identity (gsd-advisor-researcher), not the
technical spawn type.
This is a follow-up to #2555 — the SDK-side fix in that PR (#2587)
only becomes fully effective once the call sites use the right keys.
Adds `sdk/src/workflow-agent-skills-consistency.test.ts` as a
contract test: every `agent-skills <slug>` invocation in
`get-shit-done/workflows/**/*.md` must reference an existing
`agents/<slug>.md`. Fails loudly on future key typos.
Closes#2615
* test: harden workflow agent-skills regex per review feedback
Review (#2616): CodeRabbit flagged the `agent-skills <slug>` pattern
as too permissive (can match prose mentions of the string) and the
per-line scan as brittle (misses commands wrapped across lines).
- Require full `gsd-sdk query agent-skills` prefix before capture
+ `\b` around the pattern so prose references no longer match.
- Scan each file's full content (not line-by-line) so `\s+` can span
newlines; resolve 1-based line number from match index.
- Add JSDoc on helpers and on QUERY_KEY_PATTERN.
Verified: RED against base (`f30da83`) produces the same 9 violations
as before; GREEN on fixed tree.
---------
Co-authored-by: forfrossen <forfrossensvart@gmail.com>
* ci: explicit rebase check + fail-fast SDK typecheck in install-smoke
Stale-base regression guard. Root cause: GitHub's `refs/pull/N/merge`
is cached against the PR's recorded merge-base, not current main. When
main advances after a PR is opened, the cache stays stale and CI runs
against the pre-advance tree. PRs hit this whenever a type error lands
on main and gets patched shortly after (e.g. #2611 + #2622) — stale
branches replay the broken intermediate state and report confusing
downstream failures for hours.
Observed failure mode: install-smoke's "Assert gsd-sdk resolves on PATH"
step fires with "installSdkIfNeeded() regression" even when the real
cause is `npm run build` failing in sdk/ due to a TypeScript cast
mismatch already fixed on main.
Fix:
- Explicit `git merge origin/main` step in both `install-smoke.yml` and
`test.yml`. If the merge conflicts, emit a clear "rebase onto main"
diagnostic and fail early, rather than let conflicts produce unrelated
downstream errors.
- Dedicated `npm run build:sdk` typecheck step in install-smoke with a
remediation hint ("rebase onto main — the error may already be fixed
on trunk"). Fails fast with the actual tsc output instead of masking
it behind a PATH assertion.
- Drop the `|| true` on `get-shit-done-cc --claude --local` so installer
failures surface at the install step with install.js's own error
message, not at the downstream PATH assertion where the message
misleadingly blames "shim regression".
- `fetch-depth: 0` on checkout so the merge-base check has history.
* ci: address CodeRabbit — add rebase check to smoke-unpacked, fix fetch flag
Two findings from CodeRabbit's review on #2631:
1. `smoke-unpacked` job was missing the same rebase check applied to the
`smoke` job. It ran on the cached `refs/pull/N/merge` and could hit
the same stale-base failure mode the PR was designed to prevent. Added
the identical rebase-check step.
2. `git fetch origin main --depth=0` is an invalid flag — git rejects it
with "depth 0 is not a positive number". The intent was "fetch with
full depth", but the right way is just `git fetch origin main` (no
--depth). Removed the invalid flag and the `||` fallback that was
papering over the error.
* fix(#2623): resolve parent .planning root for sub_repos workspaces in SDK query dispatch
When `gsd-sdk query` is invoked from inside a `sub_repos`-listed child repo,
`projectDir` defaulted to `process.cwd()` which pointed at the child repo,
not the parent workspace that owns `.planning/`. Handlers then directly
checked `${projectDir}/.planning` and reported `project_exists: false`.
The legacy `gsd-tools.cjs` CLI does not have this gap — it calls
`findProjectRoot(cwd)` from `bin/lib/core.cjs`, which walks up from the
starting directory checking each ancestor's `.planning/config.json` for a
`sub_repos` entry that lists the starting directory's top-level segment.
This change ports that walk-up as a new `findProjectRoot` helper in
`sdk/src/query/helpers.ts` and applies it once in `cli.ts:main()` before
dispatching `query`, `run`, `init`, or `auto`. Resolution is idempotent:
if `projectDir` already owns `.planning/` (including an explicit
`--project-dir` pointing at the workspace root), the helper returns it
unchanged. The walk is capped at 10 parent levels and never crosses
`$HOME`. All filesystem errors are swallowed.
Regression coverage:
- `helpers.test.ts` — 8 unit tests covering own-`.planning` guard (#1362),
sub_repos match, nested-path match, `planning.sub_repos` shape,
heuristic fallback, unparseable config, legacy `multiRepo: true`.
- `sub-repos-root.integration.test.ts` — end-to-end baseline (reproduces
the bug without the walk-up) and fixed behavior (walk-up + dispatch of
`init.new-milestone` reports `project_exists: true` with the parent
workspace as `project_root`).
sdk vitest: 1511 pass / 24 fail (all 24 failures pre-existing on main,
baseline is 26 failing — `comm -23` against baseline produces zero new
failures). CJS: 5410 pass / 0 fail.
Closes#2623
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(#2623): remove stray .planing typo from integration test setup
Address CodeRabbit nitpick: the mkdir('.planing') call on line 23 was
dead code from a typo, with errors silently swallowed via .catch(() => {}).
The test already creates '.planning' correctly on the next line.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Codex and OpenCode install paths read `model_overrides` only from
`~/.gsd/defaults.json` (global). A per-project override set in
`.planning/config.json` — the reporter's exact setup for
`gsd-codebase-mapper` — was silently dropped, so the child agent inherited
the runtime's default model regardless of `model_overrides`.
Neither runtime has an inline `model` parameter on its spawn API
(Codex `spawn_agent(agent_type, message)`, OpenCode `task(description,
prompt, subagent_type, task_id, command)`), so the per-agent model must
reach the child via the static config GSD writes at install time. That
config was being populated from the wrong source.
Fix: add `readGsdEffectiveModelOverrides(targetDir)` which merges
`~/.gsd/defaults.json` with per-project `.planning/config.json`, with
per-project keys winning on conflict. Both install sites now call it and
walk up from the install root to locate `.planning/` — matching the
precedence `readGsdRuntimeProfileResolver` already uses for #2517.
Also update the Codex Task()->spawn_agent mapping block so it no longer
says "omit" without context: it now documents that per-agent overrides
are embedded in the agent TOML and notes the restriction that Codex
only permits `spawn_agent` when the user explicitly requested sub-agents
(do the work inline otherwise).
Regression tests (`tests/bug-2256-model-overrides-transport.test.cjs`)
cover: global-only, project-only, project-wins-on-conflict, walking up
from a nested `targetDir`, Codex TOML `model =` emission, and OpenCode
frontmatter `model:` emission.
Closes#2256
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
