The previous implementation filtered ALL .planning/-only commits, including
milestone archive commits, STATE.md, ROADMAP.md, and PROJECT.md updates.
Merging the PR branch then left the target with inconsistent planning state.
Fixes by distinguishing two categories of .planning/ commits:
- Structural (STATE.md, ROADMAP.md, MILESTONES.md, PROJECT.md,
REQUIREMENTS.md, milestones/**): INCLUDED in PR branch
- Transient (phases/, quick/, research/, threads/, todos/, debug/,
seeds/, codebase/, ui-reviews/): EXCLUDED from PR branch
The git rm in create_pr_branch is now scoped to transient subdirectories
only, so structural files survive cherry-pick into the PR branch.
Adds regression test asserting structural file handling is documented.
Closes#2004
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
When a phase completes, the offer_next step now checks whether CONTEXT.md
already exists for the next phase before presenting options.
- If CONTEXT.md is absent: /gsd-discuss-phase is the recommended first step
- If CONTEXT.md exists: /gsd-plan-phase is the recommended first step
Adds regression test asserting conditional routing is present.
Closes#2002
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
replaceInCurrentMilestone() locates content by finding the last </details>
in the ROADMAP and only operates on text after that boundary. When the
current (in-progress) milestone section is itself wrapped in a <details>
block (the standard /gsd-new-project layout), the phase section's
**Plans:** counter lives INSIDE that block. The replacement target ends up
in the empty space after the block's closing </details>, so the regex never
matches and the plan count stays at 0/N permanently.
Fix: switch the plan count update to use direct .replace() on the full
roadmapContent, consistent with the checkbox and progress table updates
that already use this pattern. The phase-scoped heading regex
(### Phase N: ...) is specific enough to avoid matching archived phases.
Adds two regression tests covering: (1) plan count updates inside a
<details>-wrapped current milestone, and (2) phase 2 plan count is not
corrupted when completing phase 1.
W006 (Phase in ROADMAP.md but no directory on disk) fired for every phase
listed in ROADMAP.md that lacked a phase directory, including future phases
that haven't been started yet. This produced false DEGRADED health status
on any project with more than one phase planned.
Fix: before emitting W006, check the ROADMAP summary list for a
'- [ ] **Phase N:**' unchecked checkbox. Phases explicitly marked as not
yet started are intentionally absent from disk -- skip W006 for them.
Phases with a checked checkbox ([x]) or with no summary entry still
trigger W006 as before.
Adds two regression tests: one verifying W006 is suppressed for unchecked
phases, and one verifying W006 still fires for checked phases with no disk
directory.
When gsd-tools commit is invoked with --files and one of the listed files
does not exist on disk, the previous code called git rm --cached which
staged and committed a deletion. This silently removed tracked planning
files (STATE.md, ROADMAP.md) from the repository whenever they were
temporarily absent on disk.
Fix: when explicit --files are provided, skip files that do not exist
rather than staging their deletion. Only the default (.planning/ staging
path) retains the git rm --cached behavior so genuinely removed planning
files are not left dangling in the index.
Adds regression tests verifying that missing files in an explicit --files
list are never staged as deletions.
* fix(hooks): skip read-guard advisory on Claude Code runtime (#1984)
Claude Code natively enforces read-before-edit at the runtime level,
so the gsd-read-guard.js advisory is redundant — it wastes ~80 tokens
per Write/Edit call and clutters tool flow with system-reminder noise.
Add early exit when CLAUDE_SESSION_ID is set (standard Claude Code
session env var). Non-Claude runtimes (OpenCode, Gemini, etc.) that
lack native read-before-edit enforcement continue to receive the
advisory as before.
Closes#1984
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(hooks): sanitize runHook env to prevent test failures in Claude Code
The runHook() test helper now blanks CLAUDE_SESSION_ID so positive-path
tests pass even when the test suite runs inside a Claude Code session.
The new skip test passes the env var explicitly via envOverrides.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
cmdPhaseComplete used replaceInCurrentMilestone() to update the overview
bullet checkbox (- [ ] → - [x]), but that function scopes replacements
to content after the last </details> tag. The current milestone's
overview bullets appear before any <details> blocks, so the replacement
never matched.
Switch to direct .replace() which correctly finds and updates the first
matching unchecked checkbox. This is safe because unchecked checkboxes
([ ]) only exist in the current milestone — archived phases have [x].
Closes#1998
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
The convertSlashCommandsToCodexSkillMentions function only converted
colon-style skill invocations (/gsd:command) but not hyphen-style
command references (/gsd-command) used in workflow output templates
(Next Up blocks, phase completion messages, etc.). This caused Codex
users to see /gsd- prefixed commands instead of $gsd- in chat output.
- Add regex to convert /gsd-command → $gsd-command with negative
lookbehind to exclude file paths (e.g. bin/gsd-tools.cjs)
- Strip /clear references in Codex output (no Codex equivalent)
- Add 5 regression tests covering command conversion, path
preservation, and /clear removal
Co-authored-by: Lakshman <lakshman@lakshman-GG9LQ90J61.local>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(core): preserve letter suffix case in normalizePhaseName (#1962)
normalizePhaseName uppercased letter suffixes (e.g., "16c" → "16C"),
causing directory/roadmap mismatches on case-sensitive filesystems.
init progress couldn't match directory "16C-name" to roadmap "16c".
Preserve original case — comparePhaseNum still uppercases for sorting
(correct), but normalizePhaseName is used for display and directory
creation where case must match the roadmap.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* test(phase): update existing test to expect preserved letter case
The 'uppercases letters' test asserted the old behavior (3a → 03A).
With normalizePhaseName now preserving case, update expectations to
match (3a → 03a) and rename the test to 'preserves letter case'.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* feat(review): add per-CLI model selection via config
- Add review.models.<cli> dynamic config keys to VALID_CONFIG_KEYS
- Update review.md to read model preferences via config-get at runtime
- Null/missing values fall back to CLI defaults (backward compatible)
- Add key suggestion for common typo (review.model)
- Update planning-config reference doc
Closes#1849
* fix(review): handle absent and null model config gracefully
Address PR #1859 review feedback from @trek-e:
1. Add `|| true` to all four config-get subshell invocations in
review.md so that an absent review.models.<cli> key does not
produce a non-zero exit from the subshell. cmdConfigGet calls
error() (process.exit(1)) when the key path is missing; the
2>/dev/null suppresses the message but the exit code was still
discarded silently. The || true makes the fall-through explicit
and survives future set -e adoption.
2. Add `&& [ "$VAR" != "null" ]` to all four if guards. cmdConfigSet
does not parse the literal 'null' as JSON null — it stores the
string 'null' — and cmdConfigGet --raw returns the literal text
'null' for that value. Without the extra guard the workflow would
pass `-m "null"` to the CLI, which crashes. The issue spec
documents null as the "fall back to CLI default" sentinel, so this
restores the contract.
3. Add tests/review-model-config.test.cjs covering all five cases
trek-e listed:
- isValidConfigKey accepts review.models.gemini (via config-set)
- isValidConfigKey accepts review.models.codex (via config-set)
- review.model is rejected and suggests review.models.<cli-name>
- config-set then config-get round-trip with a model ID
- config-set then config-get round-trip with null (returns "null")
Tests follow the node:test + node:assert/strict pattern from
tests/agent-skills.test.cjs and use runGsdTools from helpers.cjs.
Closes#1849
Backlog phases use 999.x numbering and should not be counted when
calculating the next sequential phase ID. Without this fix, having
backlog phases causes the next phase to be numbered 1000+.
Co-authored-by: gg <grgbrasil@gmail.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
cmdStateUpdateProgress, cmdStateAddDecision, cmdStateAddBlocker,
cmdStateResolveBlocker, cmdStateRecordSession, and cmdStateBeginPhase from
bare readFileSync+writeStateMd to readModifyWriteStateMd, eliminating the
TOCTOU window where two concurrent callers read the same content and the
second write clobbers the first.
Atomics.wait(), matching the pattern already used in withPlanningLock in
core.cjs.
and core.cjs and register a process.on('exit') handler to unlink them on
process exit. The exit event fires even when process.exit(1) is called
inside a locked region, eliminating stale lock files after errors.
read-modify-write body of setConfigValue in a planning lock, preventing
concurrent config-set calls from losing each other's writes.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(core): resolve @file: references in gsd-tools stdout (#1891)
Workflows used bash-specific `if [[ "$INIT" == @file:* ]]` to detect
when large JSON was written to a temp file. This syntax breaks on
PowerShell and other non-bash shells.
Intercept stdout in gsd-tools.cjs to transparently resolve @file:
references before they reach the caller, matching the existing --pick
path behavior. The bash checks in workflow files become harmless
no-ops and can be removed over time.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* docs(config): add missing config fields to planning-config.md (#1880)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Tibsfox <tibsfox@tibsfox.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Running gsd-update (re-running the installer) silently deleted two
user-generated files:
- get-shit-done/USER-PROFILE.md (created by /gsd-profile-user)
- commands/gsd/dev-preferences.md (created by /gsd-profile-user)
Root causes:
1. copyWithPathReplacement() calls fs.rmSync(destDir, {recursive:true})
before copying, wiping USER-PROFILE.md with no preserve allowlist.
2. The legacy commands/gsd/ cleanup at ~line 5211 rmSync'd the entire
directory, wiping dev-preferences.md.
3. The backup path in profile-user.md pointed to the same directory
that gets wiped, so the backup was also lost.
Fix:
- Add preserveUserArtifacts(destDir, fileNames) and restoreUserArtifacts()
helpers that save/restore listed files around destructive wipes.
- Call them in install() before the get-shit-done/ copy (preserves
USER-PROFILE.md) and before the legacy commands/gsd/ cleanup
(preserves dev-preferences.md).
- Fix profile-user.md backup path from ~/.claude/get-shit-done/USER-PROFILE.backup.md
to ~/.claude/USER-PROFILE.backup.md (outside the wiped directory).
Closes#1924
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Replaces direct fs.writeFileSync calls for STATE.md, ROADMAP.md, and
config.json with write-to-temp-then-rename so a process killed mid-write
cannot leave an unparseable truncated file. Falls back to direct write if
rename fails (e.g. cross-device). Adds regression tests for the helper.
Closes#1915
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
The requirement marking function used test() then replace() on the
same global-flag regex. test() advances lastIndex, so replace() starts
from the wrong position and can miss the first match.
Replace with direct replace() + string comparison to detect changes.
Also drop unnecessary global flag from done-check patterns that only
need existence testing, and eliminate the duplicate regex construction
for the table pattern.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Local installs wrote bare relative paths (e.g. `node .claude/hooks/...`)
into settings.json. Claude Code persists the shell's cwd between tool
calls, so a single `cd subdir` broke every hook for the rest of the
session.
Prefix all 9 local hook commands with "$CLAUDE_PROJECT_DIR"/ so path
resolution is always anchored to the project root regardless of cwd.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When --default <value> is passed, config-get returns the default value
(exit 0) instead of erroring (exit 1) when the key is absent or
config.json doesn't exist. When the key IS present, --default is
ignored and the real value returned.
This lets workflows express optional config reads without defensive
`2>/dev/null || true` boilerplate that obscures intent and is fragile
under `set -e`.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The installer writes gsd-file-manifest.json to the runtime config root
at install time but uninstall() never removed it, leaving stale metadata
after every uninstall. Add fs.rmSync for MANIFEST_NAME at the end of the
uninstall cleanup sequence.
Regression test: tests/bug-1908-uninstall-manifest.test.cjs covers both
global and local uninstall paths.
Closes#1908
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* feat(references): add common bug patterns checklist for debugger
Create a technology-agnostic reference of ~80%-coverage bug patterns
ordered by frequency — off-by-one, null access, async timing, state
management, imports, environment, data shape, strings, filesystem,
and error handling. The debugger agent now reads this checklist before
forming hypotheses, reducing the chance of overlooking common causes.
Closes#1746
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(references): use bold bullet format in bug patterns per GSD convention (#1746)
- Convert checklist items from '- [ ]' checkbox format to '- **label** —'
bold bullet format matching other GSD reference files
- Scope test to <patterns> block only so <usage> section doesn't fail
the bold-bullet assertion
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Add fs.existsSync() guards to all .js hook registrations in install.js,
matching the pattern already used for .sh hooks (#1817). When hooks/dist/
is missing from the npm package, the copy step produces no files but the
registration step previously ran unconditionally for .js hooks, causing
"PreToolUse:Bash hook error" on every tool invocation.
Each .js hook (check-update, context-monitor, prompt-guard, read-guard,
workflow-guard) now verifies the target file exists before registering
in settings.json, and emits a skip warning when the file is absent.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Fix mode: "code-first"/"plan-first"/"hybrid" → "interactive"/"yolo"
(verified against templates/config.json and workflows/new-project.md)
- Fix discuss_mode: "auto"/"analyze" → "assumptions"
(verified against workflows/settings.md line 188)
- Add regression tests asserting correct values and rejecting stale ones
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
next-decimal and insert-phase only scanned directory names in
.planning/phases/ when calculating the next available decimal number.
When agents added backlog items by writing ROADMAP.md entries and
creating directories without calling next-decimal, the function would
not see those entries and return a number that was already in use.
Both functions now union directory names AND ROADMAP.md phase headers
(e.g. ### Phase 999.3: ...) before computing max + 1. This follows the
same pattern already used by cmdPhaseComplete (lines 791-834) which
scans ROADMAP.md as a fallback for phases defined but not yet
scaffolded to disk.
Additional hardening:
- Use escapeRegex() on normalized phase names in regex construction
- Support optional project-code prefix in directory pattern matching
- Handle edge cases: missing ROADMAP.md, empty/missing phases dir,
leading-zero padded phase numbers in ROADMAP.md
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* feat(reapply-patches): post-merge verification to catch dropped hunks
Add a post-merge verification step to the reapply-patches workflow that
detects when user-modified content hunks are silently lost during
three-way merge. The verification performs line-count sanity checks and
hunk-presence verification against signature lines from each user
addition.
Warnings are advisory — the merge result is kept and the backup remains
available for manual recovery. This strengthens the never-skip invariant
from PR #1474 by ensuring not just that files are processed, but that
their content survives the merge intact.
Closes#1758
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* enhance(reapply-patches): add structural ordering test and refactor test setup (#1758)
- Add ordering test: verification section appears between merge-write
and status-report steps (positional constraint, not just substring)
- Move file reads into before() hook per project test conventions
- Update commit prefix from feat: to enhance: per contribution taxonomy
(addition to existing workflow, not new concept)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* feat(references): add gates taxonomy with 4 canonical gate types
Define pre-flight, revision, escalation, and abort gates as the
canonical validation checkpoint types used across GSD workflows.
Includes a gate matrix mapping each workflow phase to its gate type,
checked artifacts, and failure behavior. Cross-referenced from
plan-phase and execute-phase workflows.
Closes#1715
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(agents): add gates.md reference to plan-checker and verifier per approved scope (#1715)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(agents): move gates.md to required_reading blocks and add stall detection (#1715)
- Move gates.md @-reference from <role> prose into <required_reading> blocks
in gsd-plan-checker.md and gsd-verifier.md so it loads as context
- Add stall-detection to Revision Gate recovery description
- Fix /gsd-next → next for consistent workflow naming in Gate Matrix
- Update tests to verify required_reading placement and stall detection
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Move .claude to the front of the detectConfigDir search array so Claude Code
sessions always find their own GSD install first, preventing false "update
available" warnings when an older OpenCode install coexists on the same machine.
Closes#1860
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
The "files" field in package.json listed "hooks/dist" instead of "hooks",
which excluded gsd-session-state.sh, gsd-validate-commit.sh, and
gsd-phase-boundary.sh from the npm tarball. Any fresh install from the
registry produced broken shell hook registrations.
Fix: replace "hooks/dist" with "hooks" so the full hooks/ directory is
bundled, covering both the compiled .js files (in hooks/dist/) and the
.sh source hooks at the top of hooks/.
Adds regression test in tests/package-manifest.test.cjs.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(installer): deploy commands directory in local installs (#1736)
Local Claude installs now populate .claude/commands/gsd/ with command .md
files. Claude Code reads local project commands from .claude/commands/gsd/,
not .claude/skills/ — only the global ~/.claude/skills/ is used for the
skills format. The previous code deployed skills/ for both global and local
installs, causing all /gsd-* commands to return "Unknown skill" after a
local install.
Global installs continue to use skills/gsd-xxx/SKILL.md (Claude Code 2.1.88+
format). Local installs now use commands/gsd/xxx.md (the format Claude Code
reads for local project commands).
Also adds execute-phase.md to the prompt-injection scan allowlist (the
workflow grew past 50K chars, matching the existing discuss-phase.md exemption).
Closes#1736
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(installer): fix test cleanup pattern and uninstall local/global split (#1736)
Replace try/finally with t.after() in all 3 regression tests per CONTRIBUTING.md
conventions. Split the Claude Code uninstall branch on isGlobal: global removes
skills/gsd-*/ directories (with legacy commands/gsd/ cleanup), local removes
commands/gsd/ as the primary install location since #1736.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Add end-to-end regression tests confirming the installer deploys all three
.sh hooks (gsd-session-state.sh, gsd-validate-commit.sh, gsd-phase-boundary.sh)
to the target hooks/ directory alongside .js hooks.
Root cause: the hook copy loop in install.js only handled entry.endsWith('.js')
files; the else branch for non-.js files (including .sh scripts) was absent,
so .sh hooks were silently skipped. The fix (else + copyFileSync + chmod) is
already present; these tests guard against regression.
Also allowlists execute-phase.md in the prompt-injection scan — it exceeds
the 50K size threshold due to legitimate adaptive context enrichment content
added in recent releases.
Closes#1834
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(install): preserve non-array hook entries during uninstall
Uninstall filtering returned null for hook entries without a hooks
array, silently deleting user-owned entries with unexpected shapes.
Return the entry unchanged instead so only GSD hooks are removed.
* test(install): add regression test for non-array hook entry preservation (#1825)
Fix mirrored filterGsdHooks helper to match production code and add
test proving non-array hook entries survive uninstall filtering.
* docs(references): extend planning-config.md with complete field reference
Add a comprehensive field table generated from CONFIG_DEFAULTS and
VALID_CONFIG_KEYS covering all config.json fields with types, defaults,
allowed values, and descriptions. Includes field interaction notes
(auto-detection, threshold triggers) and three copy-pasteable example
configurations for common setups.
Closes#1741
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(docs): add missing sub_repos and model_overrides to config reference (#1741)
- Add sub_repos field to planning-config.md field table
- Add model_overrides field to planning-config.md field table
- Fix test namespace map to cover both missing fields
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(docs): add thinking_partner field and plan_checker alias note (#1741)
- Add features.thinking_partner to config reference documentation
- Document plan_checker as flat-key alias of workflow.plan_check
- Move file reads from describe scope into before() hooks
- Add test coverage for thinking_partner field
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* feat(commands): add /gsd-audit-fix autonomous audit-to-fix pipeline
Chains audit, classify, fix, test, commit into an autonomous pipeline. Runs an audit (currently audit-uat), classifies findings as auto-fixable vs manual-only (erring on manual when uncertain), spawns executor agents for fixable issues, runs tests after each fix, and commits atomically with finding IDs for traceability.
Supports --max N (cap fixes), --severity (filter threshold), --dry-run (classification table only), and --source (audit command). Reverts changes on test failure and continues to the next finding.
Closes#1735
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(commands): address review feedback on audit-fix command (#1735)
- Change --severity default from high to medium per approved spec
- Fix pipeline to stop on first test failure instead of continuing
- Verify gsd-tools.cjs commit usage (confirmed valid — no change needed)
- Add argument-hint for /gsd-help discoverability
- Update tests: severity default, stop-on-failure, argument-hint
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(commands): address second-round review feedback on audit-fix (#1735)
- Replace non-existent gsd-tools.cjs commit with direct git add/commit
- Scope revert to changed files only instead of git checkout -- .
- Fix argument-hint to reflect actual supported source values
- Add type: prompt to command frontmatter
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
claude --no-input was removed in Claude Code >= v2.1.81 and causes an
immediate crash ("error: unknown option '--no-input'"). The -p/--print
flag already handles non-interactive output, so --no-input is redundant.
Adds a regression test in tests/workflow-compat.test.cjs that scans all
workflow, command, and agent .md files to ensure --no-input never returns.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(tests): allowlist execute-phase.md in prompt-injection scan
execute-phase.md grew to ~51K chars after the code-review gate step
was added in #1630, tripping the 50K size heuristic in the injection
scanner. The limit is calibrated for user-supplied input — trusted
workflow source files that legitimately exceed it are allowlisted
individually, following the same pattern as discuss-phase.md.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* feat(security): improve prompt injection scanner with 4 detection layers (#1838)
- Layer 1: Unicode tag block U+E0000–U+E007F detection in strict mode (2025 supply-chain attack vector)
- Layer 2: Character-spacing obfuscation, delimiter injection (<system>/<assistant>/<user>/<human>), and long hex sequence patterns
- Layer 3: validatePromptStructure() — validates XML tag structure of agent/workflow files against known-valid tag set
- Layer 4: scanEntropyAnomalies() — Shannon entropy analysis flagging high-entropy paragraphs (>5.5 bits/char)
All layers implemented TDD (RED→GREEN): 31 new tests written first, verified failing, then implemented.
Full suite: 2559 tests, 0 failures. security.cjs: 99.6% stmt coverage.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
execute-phase.md grew to ~51K chars after the code-review gate step
was added in #1630, tripping the 50K size heuristic in the injection
scanner. The limit is calibrated for user-supplied input — trusted
workflow source files that legitimately exceed it are allowlisted
individually, following the same pattern as discuss-phase.md.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* feat(config): add execution context profiles for mode-specific agent output
* fix(config): add enum validation for context config key
Validate context values against allowed enum (dev, research, review)
in cmdConfigSet before writing to config.json, matching the pattern
used for model_profile validation. Add rejection test for invalid
context values.
* feat(tools): add global learnings store with CRUD library and CLI support
* fix(tools): address review feedback for global learnings store
- Validate learning IDs against path traversal in learningsRead, learningsDelete, and cmdLearningsDelete
- Fix total invariant in learningsCopyFromProject (total = created + skipped)
- Wrap cmdLearningsPrune in try/catch to handle invalid duration format
- Rename raw -> content in readLearningFile to avoid variable shadowing
- Add CLI integration tests for list, query, prune error, and unknown subcommand
* feat(commands): add /gsd-explore for Socratic ideation and idea routing
Open-ended exploration command that guides developers through ideas via
Socratic questioning, optionally spawns research when factual questions
surface, then routes crystallized outputs to appropriate GSD artifacts
(notes, todos, seeds, research questions, requirements, or new phases).
Conversation follows questioning.md principles — one question at a time,
contextual domain probes, natural flow. Outputs require explicit user
selection before writing.
Closes#1729
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(commands): address review feedback on explore command (#1729)
- Change allowed-tools from Agent to Task to match subagent spawn pattern
- Remove unresolved {resolved_model} placeholder from Task spawn
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* feat(commands): add external plan import command /gsd-import
Adds a new /gsd-import command for importing external plan files into
the GSD planning system with conflict detection against PROJECT.md
decisions and CONTEXT.md locked decisions.
Scoped to --from mode only (plan file import). Uses validatePath()
from security.cjs for file path validation. Surfaces all conflicts
before writing and never auto-resolves. Handles missing PROJECT.md
gracefully by skipping constraint checks.
--prd mode (PRD extraction) is noted as future work.
Closes#1731
* fix(commands): address review feedback for /gsd-import
- Add structural tests for command/workflow files (13 assertions)
- Add REQUIREMENTS.md to conflict detection context loading
- Replace security.cjs CLI invocation with inline path validation
- Move PBR naming check from blocker list to conversion step
- Add Edit to allowed-tools for ROADMAP.md/STATE.md patching
- Remove emoji from completion banner and validation message
* docs(agents): add few-shot calibration examples for plan-checker and verifier
Closes#1723
* test(agents): add structural tests for few-shot calibration examples
Validates reference file existence, frontmatter metadata, example counts,
WHY annotations on every example, agent @reference lines, and content
structure (input/output pairs, calibration gap patterns table).
When model_profile is set to "inherit" in config.json, resolveModelInternal()
now returns "inherit" immediately instead of looking it up in MODEL_PROFILES
(where it has no entry) and silently falling back to balanced.
Also adds "inherit" to the valid profile list in verify.cjs so setting it
doesn't trigger a false validation error.
Closes#1829
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
phases clear now checks for phase dirs before deleting. If any exist and
--confirm is absent, the command exits non-zero with a message showing the
count and how to proceed. Empty phases dir (nothing to delete) succeeds
without --confirm unchanged.
Updates new-milestone.md workflow to pass --confirm (intentional programmatic
caller). Updates existing new-milestone-clear-phases tests to match new API.
Closes#1826
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Before registering each .sh hook (validate-commit, session-state, phase-boundary),
check that the target file was actually copied. If the .sh file is missing (e.g.
omitted from the npm package as in v1.32.0), skip registration and emit a warning
instead of writing a broken hook entry that errors on every tool invocation.
Closes#1817
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(cli): reject help/version flags instead of silently ignoring them (#1818)
AI agents can hallucinate --help or --version on gsd-tools invocations.
Without a guard, unknown flags were silently ignored and the command
proceeded — including destructive ones like `phases clear`. Add a
pre-dispatch check in main() that errors immediately if any never-valid
flag (-h, --help, -?, --version, -v, --usage) is present in args after
global flags are stripped. Regression test covers phases clear, generate-
slug, state load, and current-timestamp with both --help and -h variants.
Closes#1818
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(agents): convert gsd-verifier required_reading to inline wiring
The thinking-model-guidance test requires inline @-reference wiring at
decision points rather than a <required_reading> block. Convert
verification-overrides.md reference from the <required_reading> block
to an inline reference inside <verification_process> alongside the
existing thinking-models-verification.md reference.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(tests): resolve conflict between thinking-model and verification-overrides tests
thinking-model-guidance.test prohibited <required_reading> entirely, but
verification-overrides.test requires gsd-verifier.md to have a
<required_reading> block for verification-overrides.md between </role>
and <project_context>. The tests were mutually exclusive.
Fix: narrow the thinking-model assertion to check that the thinking-models
reference is not *inside* a <required_reading> block (using regex extraction),
rather than asserting no <required_reading> block exists at all. Restore the
<required_reading> block in gsd-verifier.md. Both suites now pass (2345/2345).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
