* test(#1364,#1365): add decisions regression tests (fail-first proof)
Adds tests/decisions.test.cjs with:
- #1364 recall tests: parseDecisions from markdown-header + em-dash bullets
(these FAIL on pre-T1 code, proving the bug is present before the fix)
- #1365 fail-loud tests: check.decision-coverage-plan must return passed:false
for decision-shaped but 0-extracted content (FAIL pre-T1, gate silently passed)
- extractDecisions outcome enum tests (could-not-parse/none-present/parsed)
- Parser QA matrix: CRLF, unicode headings, fenced-code suppression, both bullet forms
- Boundary/threshold tests at limit-1 (0), limit (1)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(#1364,#1365): adopt markdown-sectionizer seam in decisions.cts; add fail-loud gate
#1364 — Recall: decisions.cts now uses the seam's extractTaggedBlocks and
collectSection for the markdown-header fallback path. Em-dash bullet form
(- **D-NN — title** body) is now recognised alongside the existing colon form.
#1365 — Fail-loud: adds extractDecisions() returning a typed DecisionExtraction
{ decisions, outcome } where outcome is 'parsed' | 'none-present' | 'could-not-parse'.
The blocking gate (cmdDecisionCoveragePlan) now treats could-not-parse as
passed:false with a format-mismatch reason instead of the prior silent passed:true/skip.
gap-checker runGapAnalysis surfaces 'extracted 0 of N — possible format mismatch'
for could-not-parse instead of 'No requirements or decisions to check'.
parseDecisions remains a thin delegate over extractDecisions, so all existing
callers are unaffected.
Seam adoption: stripFencedCode (seam), extractTaggedBlocks(content,'decisions') (seam),
collectSection(content, /decisions?/i, {levelBounded,stripFences}) (seam).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(#1364,#1365): tighten could-not-parse, parse-miss fail-loud, curly-quote discretion, gap-checker FIX D
FIX A: empty <decisions> scaffolds and all-prose sections no longer return
could-not-parse; outcome is none-present unless the block/section contains
a \bD- token or a parse-miss, preventing false blocks on legitimate phases.
FIX B: parseDecisionLines now tracks parse-misses (D-NN-shaped bullets that
fail both regexes); extractDecisions returns could-not-parse when parseMisses>0
even if some decisions parsed — silent drops no longer mask format errors.
FIX C: curly-quote normalization regex now includes actual U+2018/U+2019
characters so '### Claude's Discretion' (curly apostrophe) correctly yields
trackable:false (regression vs pre-T1 behavior).
FIX D: gap-checker runGapAnalysis surfaces the decision could-not-parse
format-mismatch signal independently of whether requirements items exist —
previously masked inside `if (items.length === 0)`.
Adds 14 behavioral regression tests (fail-first verified manually before fixes).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(#1365): fail-loud gate on parse-miss regardless of covered decisions
Change the `could-not-parse` guard in `cmdDecisionCoveragePlan` and
`cmdDecisionCoverageVerify` from `decisions.length === 0 && outcome ===
'could-not-parse'` to fire on `outcome === 'could-not-parse'` alone.
Previously a CONTEXT.md with a valid D-01 (covered by the plan) plus a
malformed D-02 (parse-miss) would skip the guard (length === 1), proceed
to coverage, find D-01 covered, and silently return passed:true — hiding
the D-02 parse-miss entirely.
Adds a gate-level fail-first test that places D-01 into a ## Must Haves
section (DESIGNATED_HEADINGS_RE match) so coverage of D-01 would pass on
its own, proving the only path to passed:false is the parse-miss fix.
Also adds the matching verify-side advisory assertion.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* chore(#1364,#1365): add Fixed changeset (pr:0 placeholder)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#1364): backfill changeset PR number (1386)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
The product-name-purity gate scanned CHANGELOG.md only, never the
.changeset/*.md fragments that render into it. An impure fragment passed
PR review, sat dormant, and re-introduced a forbidden parenthetical
product description at the next release — even after CHANGELOG.md had been
hand-fixed. This is the recurrence vector behind the 1.5.0 back-merge (#1379)
failure.
- Purify the two live fragments to the already-accepted forms:
- happy-finches-travel.md: "Claude Code (background dispatch …)"
-> "Claude Code; background dispatch …"
- 924-claude-flat-skill-layout.md: "Claude (`~/.claude/…`)"
-> "Claude at `~/.claude/…`"
- Extend the #1777 gate to also scan live .changeset/*.md fragments,
reusing one shared detection helper. Archived fragments never re-render
and are intentionally out of scope.
Test-only + changeset-prose change; no production behavior change.
Closes#1384
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix(#1374): surface diagnostic when configured agent skills all fail to resolve
buildAgentSkillsBlock returned '' (only ad-hoc per-path stderr warnings) when an agent configured via agent_skills had paths that all failed to resolve — missing SKILL.md, unsafe path, invalid global name, OR a malformed (non-string/non-array) value. query agent-skills --json reported skills_count>0 with an empty block and no machine-readable signal, so a fully-dropped configuration was indistinguishable from a resolved one.
Thread an optional diagnostics collector through buildAgentSkillsBlock: route every skip warning through a warn() helper (stderr + collector), flag truthy-but-malformed config values, emit an aggregate warning when configured paths resolve to zero skills, and surface the collected reasons in a new warnings[] field on the query agent-skills --json IR. Empty arrays and falsy values stay silent (skills_count is honestly 0). skills_count semantics unchanged. Docs updated for the new IR field.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#1374): backfill changeset PR number (#1376)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* feat(#1355): detect-and-warn guard for claude-code agent-teams
GSD's multi-agent orchestration can stall under claude-code's experimental
agent-teams (a subagent's completion fails to route to the orchestrator). Per
the maintainer decision, the accepted scope is a read-only detector + one
non-fatal warning — NOT the declined run_in_background/TaskOutput conversion.
- New Teams Status Module (src/teams-status.cts → gsd-core/bin/lib/teams-status.cjs):
pure resolveTeamsStatus({runtime, env}) + thin CLI cmdTeamsStatus reusing
resolveRuntime. active = strictly-truthy env flag AND runtime === 'claude'.
- Wire `gsd-tools query teams-status [--active]` (read-only; no capability
registration needed — conformance gates govern features, not query commands).
- One non-fatal warning in plan-phase.md before the first Agent spawn, gated on
`query teams-status --active`; zero behavior change on non-claude/teams-off.
- Hermeticity: clear CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS in run-tests.cjs +
SESSION_ENV_KEYS. Docs reference + CONTEXT.md glossary. Built lib gitignored.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#1355): add changeset for teams-detect guard
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* test(#1355): bump plan-phase.md workflow size baseline (+407B for teams warning)
The non-fatal agent-teams warning block added to plan-phase.md grew it
92759 → 93166 bytes, past its committed per-file baseline ratchet. The growth
is small, deliberate, and still well under the workflow tier hard cap. Regenerate
the baseline via `npm run size:baseline` (only plan-phase.md changed).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#1355): register teams-status.cjs in the inventory manifest
The new teams-status CLI module is a tracked surface; regenerate
docs/INVENTORY-MANIFEST.json (cli_modules family) via
gen-inventory-manifest.cjs --write so the inventory-manifest-sync gate passes.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix(#1356): rewrite bare ~/.claude paths in the Cursor install branch
The Cursor branch of _applyRuntimeRewrites only rewrote the trailing-slash
.claude forms, so bare ~/.claude / $HOME/.claude references survived into
installed Cursor artifacts (skills/gsd-surface, skills/gsd-graphify,
workflows/plan-phase, workflows/autonomous), tripping the post-install
"unreplaced .claude path reference(s)" audit. Same regression class as
#983/#2418/#2545 — every other affected branch was patched; cursor was missed.
- Add the three bare-form rewrites (~/.claude, $HOME/.claude, ./.claude) to the
cursor branch, mirroring cline/trae/augment/codebuddy. They run after the
slash forms (no double-replace) and use (?![\w-]) so .claude-plugin /
.claudeignore are not corrupted.
- SKILL.md content also passes through this stage, so no stage-1 converter
change is needed. Verified: 40 bare refs in the real leaking files → 0.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#1356): add changeset for Cursor bare-path rewrite fix
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix(#1348): canonicalize Codex hooks.json writes to the nested { hooks } shape
reconcileCodexHooksJsonEvent preserved whatever shape it read, so on an empty,
absent, or legacy top-level hooks.json it wrote top-level event keys
(`{ "SessionStart": [...] }`) that current Codex (deny_unknown_fields) rejects,
instead of the canonical `{ "hooks": { "SessionStart": [...] } }`.
- Lift any top-level event arrays (legacy, empty, or mixed nested+top-level)
into the nested `hooks` table, merging same-named events so no user/legacy
entry is dropped and no stray top-level event key survives. Mirrors
reconcileCursorHooksJson.
- Collapse an empty hook table back to `{}` so removal on an absent file does
not write a spurious `{ "hooks": {} }`.
- Read path still tolerates both shapes; dedup/removal unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#1348): add changeset for Codex hooks.json canonicalization
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix(#1342): scope worktree-path-guard to GSD executor runs; fail open for no-repo targets
The PreToolUse worktree-path-guard fired for any Write/Edit in any linked git
worktree, with no check for active GSD work — so Claude Code plan-mode writing
~/.claude/plans/<slug>.md from a manually-created worktree was hard-blocked.
- Gate enforcement on the GSD isolated-executor branch namespace
(^worktree-agent-[A-Za-z0-9._/-]+$, per worktree-branch-check.md #2924); the
guard is a no-op in non-GSD linked worktrees.
- Fail open when a target resolves to no git repository (e.g. ~/.claude/plans/)
instead of blocking — that is not the #260 main-repo vector. A target inside
a .git directory still blocks (git rev-parse --is-inside-git-dir).
- The #260 different-git-root hard block (escape to the main repo) is preserved.
Detached-HEAD executors no-op the gate; this is accepted because they are
fail-closed by worktree-branch-check.md (exit 42) before committing.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#1342): add changeset for worktree-path-guard scoping fix
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* test(#1342): build dot-dot traversal path portably (Windows drive-letter fix)
The traversal test built its file_path by stripping a leading slash from an
absolute externalDir and path.join-ing it after a `..` chain. On Windows the
drive letter (C:\) is not a leading slash, so it survived and path.resolve
produced an invalid doubled-drive path (C:\C:\Users\...), which resolves to no
git repo — the hook failed open (exit 0) and the test expected a block (exit 2).
Use path.relative(worktreeDir, externalTarget) + string concat so the file_path
carries literal `..` segments that resolve to externalTarget on both posix and
win32 (no drive doubling). Verified with path.win32/path.posix.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix(#1343): parse decision bullets with text before the colon
parseDecisions() silently dropped any `- **D-NN ...:**` decision bullet
whose header had freeform text (a parenthetical, em-dash, or prose) before
the `:**`, so the blocking check.decision-coverage-plan gate computed
coverage over a narrowed set and reported a false pass.
- Broaden bulletRe to tolerate a freeform run before the colon while
preserving the optional [bracket] tag capture (drives `trackable`).
- Add a parse-miss guard: a line that looks like a D-NN bullet but still
fails the regex flushes the current decision and warns instead of
vanishing — the gate-integrity floor.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#1343): add changeset for decision-coverage false-pass fix
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* test(#1343): relocate decision-parser regression into owning module test file
CI's lint-regression-test-names bans new bug-NNNN-*.test.cjs files. Move the
9 regression cases from tests/bug-1343-parsedecisions-drop.test.cjs into the
owning parser test file tests/post-planning-gaps-2493.test.cjs (which already
exercises parseDecisions) and delete the banned file.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix(#1326): stop emitting Codex agents/openai.yaml sidecars; clean up stale ones
Codex installs wrote an agents/openai.yaml sidecar under every managed gsd-*
skill dir. Recent Codex builds index both SKILL.md and the sidecar, so each
GSD skill appeared twice in autocomplete (canonical gsd-* name + humanized
display_name).
- Replace writeCodexSkillMetadataFiles / generateCodexSkillMetadataYaml with
cleanupCodexSkillMetadataSidecars: Codex-only (if isCodex), removes stale
managed gsd-*/agents/openai.yaml and prunes the now-empty agents/ dir.
- Preserve user-owned dirs (gsd-dev-preferences), non-empty agents/ dirs, and
non-gsd dirs; lstat-guard against symlinked agents/ so a delete can never
escape the skills tree; fail-open per directory.
- Codex relies on SKILL.md alone for /skills discovery.
- Update USER-GUIDE/FEATURES docs and rewrite the #774 emission tests into
cleanup tests.
Scope: the sidecar duplicate only. The separate multi-root (~/.agents/skills
shared-skills) duplicate facet is a distinct concern, not addressed here.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#1326): add changeset for Codex sidecar cleanup
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix(#1359): migrate workflows off deprecated TaskOutput to Read(outputFile)
The map-codebase and docs-update workflows collected background sub-agent
results with the deprecated Claude Code `TaskOutput` tool using `block: true`,
which has a confirmed main-session hang after the agent completes
(anthropics/claude-code#20236).
Migrate the collection steps to the upstream-recommended pattern: keep
`run_in_background=true` on the Agent spawn, then `Read` each agent's
`outputFile` (from the `async_launched` result) once it reports completion.
Completion-marker contracts and on-disk verification are unchanged, and the
non-Claude runtime fallbacks (sequential_mapping / sequential_generation) are
preserved byte-for-byte. docs-update's timeout note no longer references the
unwired `workflow.subagent_timeout` key (it kept a literal before).
Regression coverage folded into tests/subagent-timeout.test.cjs (the owning
module for background-subagent collection). Workflow size baseline regenerated
for the justified prose growth.
Refs #1355 (same latent hang surface).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#1359): add changeset for TaskOutput migration
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Delivers option (a) from the #1314 maintainer review: thread a fourth flat
scalar check_violation_fixture through the projection so a prohibition authored
at spec-phase machine-proves fail-first and greens through the deterministic
path alone — zero hand-authoring at verify time.
- src/probe-core.cts: Prohibition gains check_violation_fixture?; projectProhibitions
emits it (both kinds) ONLY for a well-formed descriptor and ONLY when non-empty
(blank/absent -> projects absent -> producer hard-gates, never a partial green).
- src/prohibition-enforcement.cts: descriptorFromProjection reads it back into
violationFixture via the same numeric-coercion-safe scalar() normalizer.
- Tests (RED-first, proven non-vacuous by reverting both src edits): CHK-02(#1346)
projection emit, CHK-08(#1346) read-back, CHK-03(D) example round-trip, the
fast-check round-trip property extended to the 4th scalar (the contract trek-e
blocked #1301 on), and a real-subprocess COMPOSE capstone greening end-to-end
through project -> descriptorFromProjection -> default prover+runner.
- Docs flipped from 'hard-gates until #1346' to 'composes end-to-end': verify-phase.md,
prohibition-probe.md, spec-phase.md authoring, ADR-550 addendum, changeset.
#1346 now tracks only the node-test causation residual.
190 affected-suite tests green; eslint + tsc clean; size baseline regenerated.
Addresses the #1314 maintainer review (trek-e):
- Major 1 (fail-OPEN): defaultProveFailFirst's node-test branch only guarded
`if (!fixture)`. A missing/typo'd/stale violationFixture made GSD_PROHIB_SUBJECT
point at a missing file; an honest negative test threw ENOENT *inside its
callback* (a failing test named distinctly from the file), which
isNonVacuousNodeTestRed accepted as proof -> a green forged from a setup crash.
Now requires fs.existsSync(path.resolve(cwd, fixture)) before spawning, symmetric
with the lint-rule path's file-result guard. Regression test pins it (RED without
the guard); a second test pins cwd-relative fixture resolution.
- Major 2 (misleading prose): the #1278 projection carries no violationFixture, so
the deterministic-locate path always hard-gates (fail-closed) until a
check_violation_fixture scalar is threaded through. verify-phase.md and
prohibition-probe.md no longer read as if the projected path produces greens;
the ADR-550 addendum records both items. Tracked as follow-up #1346.
- Documented residual: existence is necessary but not sufficient (a red caused by
the env being set vs the subject's content); recorded as a constraint, in #1346.
- Nit: stale 'NOT attested fail-first' comment -> 'NOT machine-proven fail-first'.
64 tests pass; eslint + tsc clean; changeset valid.
auto-backmerge.yml has failed since v1.4.0, freezing next's CHANGELOG.md at 1.3.1 while main reached 1.4.5. This one-time reconciliation merges main into next with -s ours (next already contains every code fix on main — hotfix fixes #866/#921,#922/#934 are present under their original commits a7b288b9d/b866b9529/74a121bb4), overlays main's CHANGELOG.md, and prunes the 52 already-released .changeset fragments main consumed at release.
next's 1.5.0-rc.5 version is preserved; net diff vs next is CHANGELOG + fragment deletions only (verified: exclude-diff empty = zero code changes).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds a region/function-scoped negative-grep idiom to the gsd-planner verification guidance plus a warn-only `validate_plan` check (`scanFileWideNegativeGateConflict`) that flags when a task's file-wide negative grep bans a construct a sibling task legitimately requires elsewhere in the same file. ReDoS-safe (linear, no RegExp on author patterns); region-scoped gates are exempt. Warn-only — never errors, never flips `valid`.
Closes#968
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* refactor(#1307): gate intel on isCapabilityActive + loop-resolver honors capability active
Part A: intel's command gate moves from config-only isIntelEnabled to the
shared isCapabilityActive('intel', cwd) — a consistency cutover (intel has
skills:[] so its tri-state collapses to the intel.enabled config leg; the gate
now flows through the resolver's precedence + runtime-aware resolution).
Part B: loop-resolver hook rendering now gates on capability state.active
(=== true, fail-closed) instead of state.enabled, so the capability config
gate is honored by the hook consumer, not just per-hook 'when'. active is now
required in the loop-resolver input types. Regression test proves a config-
disabled (active=false) capability's unconditional hook is not rendered.
Part of #1302.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore(#1307): add changeset for intel + loop-resolver active gate
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* test(1278): RED-first descriptor parity + fail-closed guards + CHK-07 byte-stability (wave 1)
- CHK-03 (RED): extend PROB-14 parity in prohibition-probe.schema.test.cjs to carry the flat
check_kind/check_target/check_rule scalars through project->write->parseMustHavesBlock; the
non-droppable check_kind-presence assertion is the load-bearing RED trigger (fails because
projectProhibitions strips check_* on the current build).
- CHK-07 (GREEN forward-guard): probe-core.test.cjs pins descriptor-less byte-stability +
dispositionForProhibition fail-closed policy, with a t.todo marker forward-locking plan 01-02.
- CHK-06 (RED): prohibition-enforcement.test.cjs asserts descriptorFromProjection export +
fail-closed on absent/partial/unknown descriptors via the projection adapter (RED until 01-03).
- No src/*.cts or .cjs edits; no new test files; lint-test-file-count clean.
* feat(1278): add optional flat-scalar check descriptor fields to Prohibition interface (wave 2)
- check_kind?/check_target?/check_rule? mirror CheckDescriptor.kind/target/rule (minus caller-attested failFirst, #1279)
- optional so existing Prohibition consumers compile unchanged
* feat(1278): project check descriptor as flat scalars in projectProhibitions (wave 2)
- emit check_kind/check_target (+ check_rule only for lint-rule with a rule) when descriptor well-formed
- under-specified/descriptor-less items project byte-identically (CHK-07); flat scalars ride existing parseMustHavesBlock continuation-KV path (no parser rewrite)
- add CHK-02 probe-core unit cases pinning the projection
- turns CHK-03 parity test GREEN; dispositionForProhibition untouched
* feat(1278): descriptorFromProjection read-back adapter feeds fail-closed locate (wave 3)
- Add descriptorFromProjection(projected) -> CheckDescriptor | null to
src/prohibition-enforcement.cts: renames the projected flat scalars
check_kind/check_target/check_rule -> {kind,target,rule?}, or null when
the descriptor is absent/non-object (no check_kind key).
- failFirst is NEVER sourced from the projection (stays caller-attested; #1279).
- rule is set only when check_rule is a non-empty string; the adapter does NOT
re-validate kind/target/rule — an under-specified descriptor reconstructs to
one the EXISTING runProhibitionEnforcement LOCATE guard rejects (located:false,
never green). The merged #1259 guard stays the single source of fail-closed truth.
- Turns the RED CHK-06 fail-closed tests (plan 01-01) GREEN end-to-end; CHK-03 /
CHK-07 stay green. CheckDescriptor type, locate guard, dispositionForProhibition,
and parseMustHavesBlock are unchanged (additive +36/-0).
* feat(1278): verify-phase locates prohibition check from projected descriptor (wave 3)
- request.check kind/target/rule sourced from projected check_kind/check_target/check_rule via descriptorFromProjection, not verifier invention (CHK-05)
- replaces the #1278 author-supplied / tracked-follow-up note with the delivered deterministic-locate behavior
- preserves fail-closed routing: absent/partial descriptor -> never green, hard-gate in both modes
- failFirst stays a verify-time caller attestation; #1279 bounds the remaining fail-first proof
* feat(1278): spec-phase captures wired-check descriptor on test-tier resolution (wave 3)
- Step 5.6 'Keep it' / verification: test path captures check_kind/check_target/check_rule, projected onto must_haves.prohibitions for verify-phase deterministic locate (CHK-04)
- SOFT capture: a test-tier prohibition without a descriptor is still allowed (no hard authoring block); stays fail-closed/flagged downstream
- --auto captures only an unambiguous descriptor, never fabricates a check path
- failFirst NOT captured at spec-phase (verify-time attestation; #1279)
- PROB-06 soft-gate + text-mode (PROB-09) behavior unchanged
* chore(1278): re-baseline workflow size for grown verify-phase + spec-phase prose (wave 3)
- spec-phase.md 28438 -> 30343 (+1905), verify-phase.md 35362 -> 36498 (+1136)
- regenerated via npm run size:baseline (no hand-picked numbers); growth is the #1278 deterministic-locate + descriptor-capture prose
- workflow-size-budget guard green (122/122)
* docs(1278): ratify optional check descriptor in dated ADR-550 addendum + type:Changed changeset
- Append dated 2026-06-15 ADR-550 addendum ratifying the D3 prohibition-item
shape extension (optional flat-scalar check_kind/check_target/check_rule)
- Document flat-scalar rationale, deterministic projection/read-back,
fail-closed on partial/invalid/absent, #1279/policy out-of-scope
- Add .changeset/1278-prohibition-check-descriptor.md (type: Changed)
* docs(1278): document optional check descriptor in prohibition-probe reference + FEATURES
- Add 'Optional wired-check descriptor (deterministic locate, #1278)' section
to the prohibition-probe reference (flat-scalar keys, projection/read-back,
fail-closed + backward-compat, failFirst stays attested)
- Add deterministic prohibition-check descriptor source entry to FEATURES.md
- No CONTEXT.md glossary change: descriptor reuses existing wired-check /
verification:test vocabulary, no new glossary term introduced
* fix(1278): pass packaging gates — changeset pr field + retired slash-form fix
- Add required pr: 1278 to changeset (lint:changeset MISSING_PR hard requirement;
plan's 'omit if unknown' was inaccurate — issue number per #1259 convention,
updated to real PR number when opened) [Rule 3 - blocking]
- Fix retired /gsd-spec-phase -> /gsd:spec-phase at verify-phase.md:83 (wave-3
prose; caught by slash-namespace invariant #3443/bug-2543, blocked CHK-09
full-suite-green) [Rule 1 - bug]
- size:baseline + INVENTORY manifest verified in-sync post-build (no diff)
* docs(1278): add check descriptor + descriptorFromProjection to CONTEXT.md prohibition glossary
* fix(1278): harden descriptorFromProjection round-trip (numeric-coercion + stray-rule) per review
- MD-01/LW-01: narrow projected scalars to primitives + String()-coerce, so a
numeric-looking check_target (parseMustHavesBlock coerces ^\d+$ to number)
reconstructs as a string and locates instead of silently un-locating; no
as-string type-lie, satisfies no-base-to-string.
- LW-02: attach rule only for the lint-rule kind (drop a stray node-test rule).
- LW-03: document the optional check_* keys in the reference Output schema.
RED->GREEN tests added in prohibition-enforcement.test.cjs.
* chore(1278): set changeset pr to 1301
* test(1278): add fast-check property for the check-descriptor round-trip + fail-closed (trek-e review)
RULESET.TESTS.property-based-testing: the projectProhibitions -> render ->
parseMustHavesBlock -> descriptorFromProjection chain is a bijective/transformation
contract. Adds 2 fc properties to tests/probe-core.property.test.cjs (no new file;
ratchet stays at 2 for probe-core):
- well-formed descriptors survive the round-trip across the full string domain
incl. the numeric-coercion case (target/rule reconstruct as strings);
- under-specified/invalid descriptors (absent / target-less / rule-less /
unknown-kind) are always fail-closed (never green, flagged, unlocated).
Stability is asserted at the descriptorFromProjection layer (the raw parse step is
intentionally lossy for numeric scalars; the shared parser is unchanged).
---------
Co-authored-by: Tom Boucher <trekkie@nomorestars.com>
- rename tests/_ff_lint_violation.test.cjs -> .cjs so node --test does not execute the lint fixture (ENOENT on intentional lib/foo.cjs); scope local plugin to it so eslint . stays green (violation -> suppressedMessages, prover still proves it)
- migrate prohibition-probe.verify-tier.test.cjs A/B to inject proveFailFirst (FF-08: attestation alone no longer greens post-#1279)
- re-baseline verify-phase.md workflow size (35362 -> 35812) after the descriptor-shape prose grew
- changeset pr: TODO -> 1279 (issue number; update to PR number at open) so changeset + docs-required gates pass
* refactor(#1306): gate graphify on isCapabilityActive (tri-state), not config-only
graphify's command gate moves from the config-only isGraphifyEnabled to the
shared isCapabilityActive('graphify', cwd) — so graphify is off unless installed
AND surfaced AND graphify.enabled. Fixes a latent claude-hardcoding in the
resolver: resolveCapabilityRuntimeState now detects the active runtime via
resolveRuntime(cwd) (GSD_RUNTIME -> config.runtime -> 'claude') so non-Claude
runtimes (Codex/Cursor) read their own surface, not ~/.claude. Hermetic
regression test proves config-on+unsurfaced -> disabled; cross-runtime test
proves GSD_RUNTIME=codex honors CODEX_HOME. Gate fails closed on every error
path. Part of #1302.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore(#1306): add changeset for graphify tri-state gate
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(#1305): add per-capability active tri-state + isCapabilityActive to the Capability State Resolver
CapabilityStateEntry gains active = enabled && configActivation, where
configActivation resolves the capability's optional activationKey via the
shared _resolveActivationValue (absent activationKey -> true). enabled stays
installed && surfaced (unchanged). Each hook's active now also cascades the
capability config gate (active && configured). Adds isCapabilityActive(capId,
cwd) — a thin convenience over resolveCapabilityRuntimeState. cmdCapabilityState
emits active per capability. No consumer cutover yet (graphify/intel: #1306/#1307;
loop-resolver: #1310). Part of #1302.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore(#1305): add changeset for capability active tri-state
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* enhance(execute): isolated-executor rejected/over-reaching run fails safe (#1292)
When an isolated (worktree) executor run is rejected — the user declines to
merge it, the orchestrator surfaces recovery for a blocked/halted plan, or the
run over-reached the requested scope — the orchestrator must no longer
default/propose recovery by editing the primary checkout (`main`). Absent an
explicit guardrail, the LLM orchestrator could improvise "continue on main",
inverting the isolation contract at the moment it matters most.
Added an ISOLATED-RUN RECOVERY — FAIL SAFE policy: default to a safe halt that
offers a fresh, narrowly-scoped worktree or inspect/discard; editing the primary
checkout requires explicit, clearly-labeled confirmation and is never the
default/proposed option.
To respect the ADR-857 phase-6 host-loop size cap on execute-phase.md (it sits
just under the pre-phase-6 baseline), the policy is delivered as an extracted
reference fragment rather than inline:
- New `execute-phase/steps/worktree-recovery-policy.md` holds the recovery policy
(the existing FAIL-CLOSED rule #48 for base/HEAD mismatches + the #1292
fail-safe guardrail). No #48 behavior change — moved verbatim.
- execute-phase.md references the fragment at the worktree-spawn recovery point,
the step-5.5 merge decision, and the stalled-agent "switch to inline execution"
menu (which for an isolated run now follows the fail-safe policy). Net effect:
execute-phase.md shrinks below its cap.
- quick.md carries the fail-safe guardrail inline at its post-return merge/discard
decision (quick.md is not size-capped).
Scoped to the recovery offer only — no automatic scope-overreach detection
(explicitly out of scope per the issue) and no new config key. Adds content
regression tests, a USER-GUIDE note, and a workflow size-baseline update.
Closes#1292
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(changeset): Changed fragment for #1292 isolated-executor fail-safe recovery
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix(#1296): align config docs/prompts/schema with consumers
The user-facing config surface disagreed with what the consumers actually do
(subset of the #1216 audit). No runtime consumption behavior changes.
- workflow.subagent_timeout: settings-advanced.md prompt + docs/CONFIGURATION.md
said "seconds (default 600)" but the consumer (map-codebase.md) uses
milliseconds (default 300000). Relabeled all four spots in settings-advanced.md
(prompt, parse-default list, example, confirmation table) + the CONFIGURATION.md
row.
- review.models.<cli>: settings-integrations.md, docs/CONFIGURATION.md (Integration
Settings), and docs/CLI-TOOLS.md documented a shell command, but review.md injects
the value into a --model/-m flag. Relabeled to a bare model id and reconciled the
contradictory CONFIGURATION.md sections.
- workflow.test_command + workflow.build_command: consumed via config-get
(test_command in verify-phase/execute-phase/audit-fix/post-merge-gate;
build_command in post-merge-gate) and documented, but absent from validKeys so
`config set` rejected them. Registered both in config-schema.manifest.json and
documented them in references/planning-config.md (overview + complete reference).
Regression tests: behavioral config-set tests (tests/config.test.cjs) + doc-parity
content guards (tests/config-field-docs.test.cjs).
Deferred to other #1216 clusters: security-gate wiring, search_gitignored wiring,
mvp_mode, source_grounding_authority labeling, and config-set enum enforcement.
Closes#1296
Refs #1216
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(changeset): Fixed fragment for #1296 config-surface alignment
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Adds mcp__perplexity__* to both researcher profiles (generated source-of-truth) and regenerates the agents; adds a generative dispatch-table↔tools parity guard so future provider drift fails CI. Regenerates the agent-size baseline for the +20-byte frontmatter growth.
Fixes#1284
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* enh(#966): gate behavior-dependent truths on behavioral evidence in gsd-verifier
Introduce a per-truth PRESENT_BEHAVIOR_UNVERIFIED state for must-haves that assert a state transition or a cancellation/cleanup/ordering invariant whose only evidence is symbol presence + wiring. Such truths are excluded from the verified_truths score, reported as a behavior_unverified count, recorded in an always-on behavior_unverified_items frontmatter list, and routed to the existing human_needed sink — so a clean N/N can no longer be reached on symbol presence alone.
The overall-status vocabulary and the src/verification.cts seam are unchanged (the new state is per-truth only); gaps_found keeps decision-tree precedence; override-passed truths still count toward verified_truths. Mirrors the calibration into the shipped verify-phase.md workflow (with an infra/foundation carve-out), the VERIFICATION.md templates, and docs (planning-artifacts.md, AGENTS.md). gsd-verifier.md kept under its 48KB LARGE cap; size baselines regenerated.
Closes#966
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#966): add changeset fragment for gsd-verifier behavior-unverified calibration
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Maintainer CHANGES_REQUESTED (reviewed e08667e5, pre-portability-fix):
- B1: lint-rule no longer greens an unparseable target (eslintHasFatalError -> fail
closed on any fatal/parse error) or an inline-suppressed violation (eslintJsonHasRule
now scans suppressedMessages too). RED-first + real-runner repros.
- B2: both child spawns get a bounded timeout (30s node / 60s eslint) + 16MiB maxBuffer;
timeout fails closed. Injectable timeoutMs (positive-only — 0/negative can't disable
the bound) enables a fast 1.5s hang test.
- M1: scoped verify-phase.md — the check descriptor is author-supplied for now; filed
#1278 for deterministic auto-locate of the descriptor (the locate half).
- M2: added tests/prohibition-enforcement.property.test.cjs (fast-check fail-closed
invariants; within the <=2-file budget).
- m1: tapTestNames excludes # SKIP/# TODO; parseNodeTestSummary tracks # cancelled;
isNonVacuousNodeTestPass requires cancelled===0.
- m2: scoped the determinism claim to the decision/parse layer (real runner is env-dependent).
- m3: filed #1279 for machine-proven fail-first (violation-fixture probe).
- n1: -- before target in both arg builders (option-injection). n2: dropped dead token.
- B3 (Windows npx) was already fixed in 2af76306 (pushed ~65s after the review).
Verified node 22 + 24; size baseline regenerated for the verify-phase note.
Adversarial pre-submission review found the injected-runCheck tests masked a
non-functional real runner. Fixes:
- BL-01 (false green on vacuous test): the node-test runner now parses the TAP
summary and requires a NON-VACUOUS pass (>=1 test, >=1 pass, 0 fail) AND a
reported test named distinctly from the file — node --test counts an empty file
as one passing test, so counts alone could not catch it.
- SF-01 (lint anchor never greened): the lint-rule runner now runs the project
eslint as --format json and filters by ruleId, so plugin rules (local/*) load
via the flat config — bare --rule cannot load a plugin. local/no-source-grep
now genuinely greens (covered by a real, non-injected test).
- BL-02 (tautological fail-first): the runner no longer echoes the caller's
failFirst as if confirmed. failFirst is documented as caller-ATTESTED; the
producer requires attestation + a genuine non-vacuous pass. Machine-proven
fail-first (needs a violation fixture) is flagged as a tracked follow-up in
ADR-550, the changeset, FEATURES, the reference doc, and verify-phase.
- SF-02: added real-runner end-to-end tests (no injected runCheck) + pure,
exported parse/filter helpers (parseNodeTestSummary, tapTestNames,
eslintJsonHasRule, eslintFileResultCount) so the shipping branches are
mutation-pinned.
- NIT-01/02: LOCATE guard rejects empty-string rule and unknown kinds.
- Hardening: spawn checks with NODE_TEST_CONTEXT/NODE_OPTIONS scrubbed so an
ambient test-runner context cannot corrupt a verify-time result.
- Docs reconciled to the shipped behavior (no 'confirms fail-first' overclaim).
* feat(#1243): consume Claude plugin-provided skills via native Skill-tool directive + grant Skill to agent_skills-consumer agents
- Relax global skill name validation to accept namespaced form `^[A-Za-z0-9_-]+(:[A-Za-z0-9_-]+)*$`
- Namespaced names (containing colon) on claude runtime emit a Skill-tool load directive instead of a @-include line
- Namespaced names on non-claude runtimes are skipped with a warning
- Bare unresolved names retain existing warn-and-skip behavior (no promotion to directive)
- Grant `Skill` tool to all 22 agent_skills consumer agents; 5 generated agents updated via research-profiles.cjs + regen, 17 hand-authored agents edited directly
- Add 16 TDD tests in describe('bug #1243') covering happy/mixed/precedence/negative/cross-runtime/regression/grant cases
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* docs(#1243): document plugin-provided skills in agent_skills
Update the Agent Skills Injection reference in CONFIGURATION.md with
the three entry forms (project-relative, global:<name>,
global:<plugin>:<skill>), the Claude-only runtime behaviour of the
namespaced form and the warn-skip on other runtimes, the plugin
pre-install prerequisite, and the consumer-agent Skill tool grant.
Add docs/how-to/attach-a-plugin-skill-to-a-gsd-agent.md with a
step-by-step guide for installing the plugin, locating the namespaced
skill name, wiring it into agent_skills, and verifying injection.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(#1243): align agent_skills docs with emitted block format + mixed-block regression test (code-review)
- Replace two-section mixed-block example (bogus "Load these plugin-provided skills using the Skill tool:" header) with the actual single-section inline format in CONFIGURATION.md and docs/how-to/attach-a-plugin-skill-to-a-gsd-agent.md
- Fix quoted warning text in how-to doc to exactly match the emitted string: [agent-skills] WARNING: Plugin-namespaced skill "global:<name>" requires a Skill-tool-capable runtime (claude) — skipping on runtime "<runtime>"
- Replace phantom agent slugs (gsd-checker, gsd-researcher, gsd-advisor, gsd-synthesizer) in CONFIGURATION.md Supported Agent Types with real agents/gsd-*.md examples (gsd-plan-checker, gsd-phase-researcher, gsd-code-reviewer, gsd-ui-auditor, gsd-research-synthesizer)
- Add byte-identical mixed-block regression test: one path-resolvable global skill + one plugin-namespaced skill on claude runtime → asserts r.ir.block === single-section interleaved block, no secondary header
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* chore(#1243): regenerate agent-size baseline for the Skill-tool grant
The 22 agent_skills-consumer agents each grew +7 bytes from adding `Skill`
to their tools list; refresh the committed per-agent size baseline (#1074 guard).
* chore(#1243): add Added changeset fragment
* fix(#1243): traceable allow-test-rule ref + separator-agnostic byte-identical tests (CI)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(#1257): update pipe-table Status/Phase/Plan cells in planned-phase + begin-phase
cmdStatePlannedPhase ran its body-field replacements on the full file content,
so the case-insensitive ^Status: pattern matched the YAML frontmatter `status:`
line before the body `| Status | … |` cell — the cell never advanced to
'Ready to execute' and syncStateFrontmatter re-derived the stale 'planning'
status (the #1230 delta heuristic preserved it). cmdStateBeginPhase had
pipe-table else-branches only for Status/Last activity (#1256), so the Current
Position `| Phase |` / `| Plan |` cells were left stale while a spurious inline
`Phase: N — EXECUTING` line was prepended.
Both handlers now strip frontmatter before body-field replacement and update the
pipe-table cells in place via stateReplaceField, matching inline-format
behaviour. Systemic residual of #1255 / #1256.
Adds 4 regression tests (#1257 block in tests/state.test.cjs) covering both
findings — RED before the fix, GREEN after; full state-area suite (437) and
local unit suite stay green.
Closes#1257
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#1257): backfill changeset pr number (#1260)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix(#1255): advance frontmatter status for pipe-table STATE.md on begin/complete-phase
state begin-phase/complete-phase called stateReplaceField on the FULL file content, so its case-insensitive ^Status: plain-pattern matched the YAML frontmatter status: line first (no g flag) and never updated the body pipe-table Status cell; syncStateFrontmatter then re-derived the stale status from the unchanged body, freezing the frontmatter status. Fix: strip frontmatter before the body-field replacements (operate on body only), reassemble with frontmatter preserved, so the body Status cell updates and the frontmatter derives correctly — for inline AND pipe-table body formats. Also corrects the Current Position pipe-table else-branches (Status/Phase/Last-activity) to write bare, consistent cell values. Pipe-table Status is a supported body format (not rewritten to inline).
Closes#1255
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* chore(#1255): add changeset for pipe-table state status fix (#1256)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* test(#1255): fold pipe-table regression into state.test.cjs + Windows-portable frontmatter regex
Per the 2026-06 audit, new tests/bug-NNNN-*.test.cjs files are banned (lint-regression-test-names) — folded the 7 #1255 regressions into tests/state.test.cjs and removed the standalone file + its lint-test-file-count allowlist entry. Also fixed the frontmatter assertions' /^---\n/ anchors to /^---\r?\n/ (windows-test-parity-guard frontmatterAnchorLiteralNewline).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Edge-probe now surfaces a zero-classification requirement (non-empty prose, no
shape cue matched, no `shapes` override) as a single soft `unclassified — review
manually` candidate instead of silently dropping it — the exact blind spot the
probe exists to catch. Dismissible like any edge; the `shapes: []` opt-out stays
silent; `TAXONOMY` (the closed 8 categories) is unchanged. Under `--auto` the
candidate is left `unresolved`, never auto-`backstop` (a missing shape is not
evidence an edge exists).
Closes#1110
Adds the spec-time prohibition probe (spec-phase Step 5.6) — the second adapter of the probe-core resolution model. Surfaces unwritten must-NOT constraints as negative SPEC acceptance criteria with test/judgment verification tiers; fail-closed at verify time. Per ADR-550. Closes#644.
