* test(3309): red — workflow.human_verify_mode contract
New behavioral test file covers:
- workflow.human_verify_mode is a recognized config key (VALID_CONFIG_KEYS)
- defaults to 'mid-flight' (preserves current behavior)
- config-set / config-get round-trips for both values
- persists in config.json as string
- planner agent file references the flag with canonical wording, couples
end-of-phase mode with the rule that checkpoint:human-verify is not
emitted, and documents the <verify><human-check> deferred-item shape
- verifier agent file references harvesting <verify><human-check> blocks
- references/checkpoints.md documents the cost-control alternative
Source-text assertions on agent .md files are exempted via
allow-test-rule: source-text-is-the-product — those files ARE the
runtime contract loaded by AI runtimes, so asserting their wording is
the only way to verify the agents will respect the flag.
Fails 10/11 against current source. Will pass after the fix.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(3309): add workflow.human_verify_mode = end-of-phase opt-out
Each mid-flight checkpoint:human-verify halt costs a full executor
cold-start (CLAUDE.md, MEMORY.md, STATE.md, plan re-read on every
respawn) because subagent context is discarded across the pause. A plan
with N human-verify checkpoints pays the cold-start cost N+1 times. The
reporter (rentanything-nb) measured this at "tens of thousands of tokens"
per round-trip and "hundreds of thousands per week."
This adds workflow.human_verify_mode (default 'mid-flight') with an
'end-of-phase' value that:
- instructs gsd-planner to NOT emit <task type="checkpoint:human-verify">
tasks; verification details go into a <verify><human-check> sub-block
on the relevant auto task instead
- instructs gsd-verifier (Step 8) to harvest those <verify><human-check>
blocks at end-of-phase and merge them into its own human-verification
list
- the existing human_needed → HUMAN-UAT.md flow in execute-phase.md is
the single sink — no new file/writer is created
checkpoint:decision and checkpoint:human-action are unaffected — those
gate the work itself, not post-hoc verification.
Surfaces touched:
- bin/lib/config-schema.cjs, bin/lib/config.cjs — register key + default
- sdk/src/config.ts, sdk/src/query/config-schema.ts — SDK parity
- agents/gsd-planner.md — slim Detection section + reference link
- agents/gsd-verifier.md — Step 8 harvest instruction
- get-shit-done/references/planner-human-verify-mode.md — full rules,
loaded conditionally to keep planner.md under its size budget
- get-shit-done/references/checkpoints.md — surface the alternative
- docs/CONFIGURATION.md — config table row
- docs/INVENTORY.md, docs/INVENTORY-MANIFEST.json — track new reference
Tag name <human-check> chosen instead of <human> to avoid the
prompt-injection scan pattern that flags <system|assistant|human> tags.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore(3309): align changeset pr: to actual PR number
The pr: field was authored as 3319 (a guess at the next number) before
the PR was opened. Actual PR is #3325.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(3309): flip workflow.human_verify_mode default to end-of-phase
Per maintainer direction on PR #3325, end-of-phase is the new project
default. Mid-flight checkpoint:human-verify halts cost a full executor
cold-start (CLAUDE.md, MEMORY.md, STATE.md, plan re-read on respawn) per
round-trip — reported at "tens of thousands of tokens" per round-trip,
"hundreds of thousands per week" on real projects. The cost-control
mode is what new projects should get out of the box.
mid-flight remains a one-line opt-back-in via:
gsd config-set workflow.human_verify_mode mid-flight
Behavior change for existing projects: the new default takes effect
when .planning/config.json is rewritten (config-set, fresh project).
Existing in-flight PLAN.md files with checkpoint:human-verify tasks
continue to work in either mode — the flag only changes what the
planner emits next time it runs.
Surfaces updated:
- bin/lib/config.cjs, sdk/src/config.ts — default flipped
- sdk/src/config.ts docstring — describes new default + opt-back-in
- agents/gsd-planner.md — Detection section explains new default
- references/planner-human-verify-mode.md — reordered modes; added
guidance on when to opt back into mid-flight
- references/checkpoints.md — surface the default flip and the why
- docs/CONFIGURATION.md — table row reflects new default + reason
- tests/feat-3309-human-verify-mode.test.cjs — default test asserts
end-of-phase
- .changeset/fierce-geese-march.md — describes the default flip and
the migration semantics
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix: address human verify mode review
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(3317): red — SDK detect-custom-files must scan skills/
Mirrors tests/bug-2942-detect-custom-skills.test.cjs on the SDK side.
The SDK's GSD_MANAGED_DIRS array omits 'skills', so user-added skills
under <config-dir>/skills/<name>/ are never returned and get destroyed
on /gsd-update. New vitest covers:
- detects custom skill at skills/<name>/SKILL.md
- does not flag manifest-tracked skill as custom
- still detects custom files under get-shit-done/workflows/ (regression)
- custom_count matches custom_files.length across multiple skills
Fails 2/4 against current SDK source. Will pass after fix.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(3317): SDK detect-custom-files now scans skills/ (parity with CJS)
The SDK port of detect-custom-files declared GSD_MANAGED_DIRS without
'skills', while the canonical bin/gsd-tools.cjs port (which the SDK
docstring explicitly cites as its source) had the entry. Because
update.md prefers gsd-sdk over the CJS shim, real-world users with the
SDK installed never had their custom skills detected — the installer's
'Installed N skills to skills/' step then wiped any non-manifest skill
without backing it up to gsd-user-files-backup/. Real-world incident:
skills/gsd-roadmap/SKILL.md (fully user-owned) destroyed during the
1.40.0 → 1.41.0 update, recoverable only via Time Machine.
One-line fix adds 'skills' to the SDK's GSD_MANAGED_DIRS, matching the
CJS source the port was supposed to mirror. The 4 vitest cases added in
the prior commit now all pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore: correct changeset pr number
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix: reconcile planner action contract
The deep_work_rules workflow block was stronger than the planner agent contract: it required verbatim context copies and self-sufficient action text, so planners were incentivized to inline implementation code. Bound action content to directive prose with concrete identifiers, allow behavior/test acceptance criteria, and pin the cross-file contract with a regression test. Closes#3320.
* chore: add changeset for planner contract fix
Synthesizes 17 recurring failure modes observed across PRs #3306–#3325
plus sibling fixes (#3240, #3242, #3245, #3257, #3261, #3267, #3286,
#3287). Appends a new machine-oriented section using the same
backtick-keyed `KEY.SUB-KEY=value` format as the existing AI Ops Memory
and Release Notes Standard sections.
Each anti-pattern carries `symptom`, `examples`, `detect`, and
`fix-forward` sub-keys so AI agents can grep deterministically and apply
the remediation without re-deriving the analysis.
Closes#3327
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* refactor: tighten sdk-first architecture seams
Refs #3312
* refactor: finish state document seam cleanup
Refs #3312
* test: harden minimal install cleanup assertion
* ci: support sdk-scoped package lock
* fix(3316): restore root package-lock.json and align changeset pr ref
Reverts dec57a83 ("ci: support sdk-scoped package lock") and restores
the root package-lock.json that c249d34d deleted. The deletion was the
wrong direction:
- The root package.json declares its own runtime and dev deps
(@anthropic-ai/claude-agent-sdk, ws, c8). Without a root lockfile,
`npm install --no-package-lock` resolves whatever satisfies semver at
install time — CI today and CI in six months can install different
transitive trees, defeating reproducibility.
- The lockfile has been part of every release on this repo (long
history on main); removing it loses the npm audit / Dependabot
target without compensating benefit.
- The CI workaround pattern (cache-dependency-path: sdk/package-lock.json
+ `npm install --no-package-lock`) papered over the symptom rather
than fix the cause.
Also fix the changeset pr: from 3312 (issue) to 3316 (PR). CONTEXT.md
flags this exact failure mode as a recurring CodeRabbit finding.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix: address coderabbit review findings
* fix: close remaining coderabbit threads
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Closes#3310
Wires `ERROR_REASON.SDK_UNKNOWN_COMMAND` and `ERROR_REASON.USAGE` into the remaining untyped error paths in `gsd-tools.cjs` (template, frontmatter, requirements, milestone, uat, todo, workstream, graphify, learnings subcommand routers and `--cwd`/missing-required-arg paths). Tests assert via `JSON.parse(stderr).reason` rather than substring matching on prose. Closure regression guard locks the canonical `{ok, reason, message}` shape for every newly-typed path.
Also moves `--json-errors` activation up to the top of `main()` so `--cwd` validation paths emit JSON rather than plain text when `GSD_JSON_ERRORS=1` is set.
* chore: revert redundant CHANGELOG.md row from #3308
PR #3308 added a Shared scanPhasePlans helper row to CHANGELOG.md while
also dropping the canonical .changeset/3262-extract-scan-phase-plans.md
fragment. CONTRIBUTING.md (line 110) prohibits hand-editing CHANGELOG.md
since the release workflow folds .changeset/*.md fragments into the
file at release time — the manual row would duplicate at next release.
Removes only the 4-line Enhancement block added by #3308. The fragment
remains unchanged and is the single source of truth for this entry.
Refs #3262
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore(changeset): add Removed fragment for redundant CHANGELOG row cleanup
Documents the deletion in PR #3313 of the 4-line ### Enhancement block
that #3308 hand-wrote into CHANGELOG.md alongside its canonical
.changeset/3262-extract-scan-phase-plans.md fragment.
Refs #3262
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 pull_request_review event path is held in `action_required` whenever
the triggering_actor is an outside collaborator — exactly the case the
workflow exists to handle. The blocked reviewer's own action is what
gates the dismissal that would remove them, so the run never executes.
Add `schedule: */15 * * * *` and `workflow_dispatch` triggers. Scheduled
runs execute as github-actions[bot], bypassing the gate. The script
branches on context.eventName: the review event keeps the single-review
fast path; schedule/dispatch paginates open PRs and reviews, applying
the same role/blocklist check across all APPROVED reviews.
resolveRole() now caches per-login lookups so the schedule path doesn't
re-query the same reviewer once per PR. Concurrency group falls back to
'scheduled' for non-PR events so polls serialize.
Fixes#3314
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(phase-dir): add red test for k015 prefix-drift in plan-milestone-gaps and import workflows (#3298)
Asserts that:
- plan-milestone-gaps.md step 8 does not use bare {NN}-{name} mkdir pattern
- plan-milestone-gaps.md step 8 uses phase.add or expected_phase_dir
- import.md plan_convert does not use bare {NN}-{slug} mkdir pattern
- import.md plan_convert uses expected_phase_dir from init.phase-op
These 5 tests are RED until the fix lands.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(phase-dir): add projectCode prefix to phase-dir construction in plan-milestone-gaps and import workflows (#3298)
Both plan-milestone-gaps.md step 8 and import.md plan_convert step were
constructing phase directories using raw {NN}-{name}/{NN}-{slug} template
patterns, bypassing the project_code prefix from .planning/config.json.
Fix: both steps now call `gsd-sdk query init.phase-op <N>` and consume the
`expected_phase_dir` field (which includes the `<CODE>-<NN>-<slug>` prefix
when project_code is set), matching the pattern already established by
PR #3292 for /gsd-discuss-phase and /gsd-plan-phase.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(phase-dir): apply project_code prefix to backlog phase dir in add-backlog workflow (k015 sibling, #3298)
Sibling k015 audit found a third drift site: add-backlog.md step 4 was
constructing the 999.x backlog phase directory using raw ${NEXT}-${SLUG}
without applying the project_code prefix from .planning/config.json.
Fix: read project_code via `gsd-sdk query config-get project_code --raw`
and prepend `${CODE}-` when set, matching the pattern used by phase.insert
(which already applies project_code to decimal phases in phase.cjs line 736).
Also extends bug-3298 regression test to cover this third site.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* chore(changeset): add changeset for #3298 phase-dir prefix drift fix
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* chore(changeset): set pr: 3306 in changeset for #3298; resolve stash conflict in live-command-registry.cjs
The conflict was cosmetic (string concat → template literals) introduced by
accidental git stash during test verification. Taking the newer template-literal
form throughout.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
