docs-update Skill | Agent Skills

Codex compatibility note:

Invoke repository skills with $skill-name in Codex; this mirrored copy rewrites legacy Claude /skill-name references.

Task tracker mandate: BEFORE executing any workflow or skill step, create/update task tracking for all steps and keep it synchronized as progress changes.

User-question prompts mean to ask the user directly in Codex.

Ignore Claude-specific mode-switch instructions when they appear.

Strict execution contract: when a user explicitly invokes a skill, execute that skill protocol as written.

Subagent authorization: when a skill is user-invoked or AI-detected and its protocol requires subagents, that skill activation authorizes use of the required spawn_agent subagent(s) for that task.

Do not skip, reorder, or merge protocol steps unless the user explicitly approves the deviation first.

For workflow skills, execute each listed child-skill step explicitly and report step-by-step evidence.

If a required step/tool cannot run in this environment, stop and ask the user before adapting.

Codex Project-Reference Loading (No Hooks)

Codex uses static project-reference loading instead of runtime-injected project docs. When coding, planning, debugging, testing, or reviewing, open project docs explicitly using this routing.

Always read:

docs/project-config.json (project-specific paths, commands, modules, and workflow/test settings)
docs/project-reference/docs-index-reference.md (routes to the full docs/project-reference/* catalog)
docs/project-reference/lessons.md (always-on guardrails and anti-patterns)

Missing/stale context route: If docs/project-config.json, the docs index, lessons.md, CLAUDE.md, AGENTS.md, or any task-required reference doc is missing or stale, auto-run $project-init or the narrow setup route ($project-config, $docs-init, $scan-all, $scan --target=<key>, $claude-md-init) before ordinary project-specific work. If Codex mirrors or AGENTS.md are missing/stale, ask the user to run $sync-codex; do not auto-run it.

Situation-based docs:

Project structure/architecture/tech-stack/deployment/setup (any layer — backend, frontend, or infra): project-structure-reference.md
Backend/CQRS/API/domain/entity changes: backend-patterns-reference.md, domain-entities-reference.md
Frontend/UI/styling/design-system: frontend-patterns-reference.md, scss-styling-guide.md, design-system/README.md
Spec authoring, docs/specs/ pathing, or TC format: feature-spec-reference.md, spec-system-reference.md, spec-principles.md
Behavior/public-contract changes or spec-test-code sync: workflow-spec-test-code-cycle-reference.md plus the spec docs above
Derived spec indexes/ERDs/reimplementation guides: spec-system-reference.md and source Feature Specs under docs/specs/
Integration test implementation/review: integration-test-reference.md
E2E test implementation/review: e2e-test-reference.md
Code review/audit work: code-review-rules.md plus domain docs above based on changed files

Do not read all docs blindly. Start from docs-index-reference.md, then open only relevant files for the task.

[BLOCKING] Execute skill steps in declared order. NEVER skip, reorder, or merge steps without explicit user approval. [BLOCKING] Before each step or sub-skill call, update task tracking: set in_progress when step starts, set completed when step ends. [BLOCKING] Every completed/skipped step MUST include brief evidence or explicit skip reason. [BLOCKING] If Task tools are unavailable, create and maintain an equivalent step-by-step plan tracker with the same status transitions.

Quick Summary

Goal: Detect impacted docs from code changes and orchestrate updates across all doc types so every code/spec/test change leaves documentation in sync — impacted project-reference docs, project-config.json sections, Feature Specs, §8 TCs, test-code links, and derived indexes all reflect the shipped behavior, with zero drift left silent.

Summary:

This skill is a ROUTER, not an author — start with Phase 0 triage (git diff → categorize → dedup modules → check existing docs) and delegate each doc type to its owner ($spec, $spec [mode=tests], $spec [mode=sync], $spec-index, $tech-spec for derived technical views); NEVER write §8, docs/specs/, or derived technical spec content directly. — why: dual authorship diverges spec from index/view.
Main steps (each impact-gated; skipped phase → mark completed with reason): Phase 0 triage (git diff → categorize → dedup modules → record existing-doc state) → Phase 1 project context sync (impact-map → PARALLEL verify of the impacted docs/project-reference/** docs + docs/project-config.json sections + README.md) → Phase 2 $spec (§1–§7 Feature Spec; doc-first BLOCK when feature behavior changed but no Spec exists) → Phase 2.5 $spec-index (derived bucket INDEX/ERD refresh, optional) → Phase 2.6 $tech-spec (derived technical view refresh/audit, optional when technical tree is affected) → Phase 3 $spec [mode=tests] (§8 TCs) → Phase 4 $spec [mode=sync] (§8 ↔ test code) → Phase 5 summary report → final review (#8 runs the Step 2.4 code↔spec sync-verify).
Create ALL 8 tasks via task tracking before touching any file; run the fixed phase order 0 → 1 → 2 → 2.5/2.6 → 3 → 4 → 5 → final review — fast-exit is a decision, never a silent omission.
The final pass (Step 2.4) is the workflow's last gate: per touched module verify shipped code against §3 ACs, §4 BRs, §8 TCs — a removed/weakened [HARD] BR is a code-vs-spec contradiction that BLOCKS completion.
Output is tech-agnostic prose (no framework/product names outside evidence fields) and traceability-first (update FR-/BR-/OP-/TC- logical IDs before prose); ALWAYS write the Phase 5 summary report as the audit trail.

Orchestration Model:

git diff → Triage → Phase 1: Project Context Sync (PARALLEL, impact-scoped)
                  │            ├─ impacted docs/project-reference/** — verify → patch → (escalate to $scan --target=X)
                  │            ├─ impacted docs/project-config.json sections — verify → merge → validate
                  │            └─ README.md / project docs (docs-manager)
                  → Phase 2: $spec (business feature docs)
                  → Phase 2.5: $spec-index (derived index/ERD refresh) [optional]
                  → Phase 2.6: $tech-spec (derived technical view refresh/audit) [optional]
                  → Phase 3: $spec [mode=tests] (§8 test specifications)
                  → Phase 4: $spec [mode=sync] (§8 ↔ test code sync)
                  → Phase 5: Summary Report

Key Rules:

Router only — NEVER duplicate sub-skill logic or write Section 8 / docs/specs/ content
[BLOCKING] Freshness is impact-scoped, never assumed. Phase 1 verifies only the docs/project-reference/** docs and docs/project-config.json sections the diff can actually rot (routed by node .claude/scripts/doc-impact-map.cjs), and reports a per-doc verdict FRESH | PATCHED | RESCAN REQUIRED | UNVERIFIED. A doc nobody checked is UNVERIFIED — NEVER FRESH. — why: these docs are injected into every downstream AI context, so a silent stale line teaches every later agent a codebase that no longer exists.
[BLOCKING] An impact-scoped verify NEVER writes . Only a full $scan --target=X may move that stamp; the narrow pass writes  instead — and only in a doc that already carries a Last scanned stamp (Step 1.6). — why: Last scanned drives the 60-day full-rescan gate (.claude/hooks/lib/session-init-helpers.cjs:769); resetting it from a partial check would buy speed by disabling the very net that catches whole-doc rot.
**[BLOCKING] A PATCHED docs/project-reference/**doc MUST run$prompt-enhance <doc> before its verdict is final** (Step 1.3) — keeps the doc as concise as possible while staying valuable enough for AI; skip only for a stamp/count-only edit.
Phase 1/docs-manager MUST NOT own any docs/specs/**, test-spec, spec-index/ERD, or derived technical-view path.
Every excluded artifact is explicitly reserved to its child skill ($spec, $spec-index, or $tech-spec) so one canonical writer owns it.
Exclude docs/specs/** and generated technical views from every docs-manager brief and write set.
Each phase checks whether needed before invoking — skip phases with no impact
Step-to-skill order is fixed — run phases sequentially, never out of order
ALWAYS report what was checked, even if nothing needed updating
Pass triage context (changed files, detected modules, impacted sections) to each sub-skill via $ARGUMENTS
MUST ATTENTION dedup module list — backend + frontend changes for same module = ONE entry
MUST ATTENTION track step state live: in_progress -> execute -> completed (or completed with skip reason)
For .claude skills/hooks/workflows/sync tooling changes, flag generated mirror sync status (npm run codex:sync completed or explicit N/A). docs-update routes and reports this check; it does not edit generated mirrors directly.
[BLOCKING] Tech-agnostic output: when updating spec/specs/README/INDEX, do NOT introduce framework/product/language/design-pattern names into prose or headings — preserve the evidence-field exception (**Evidence**, CoveredBy, legacy IntegrationTest, [Source:], frontmatter, Mermaid). Authority: docs/project-reference/spec-principles.md §3.
[BLOCKING] M3 Traceability Update: See .claude/skills/shared/sdd-artifact-contract.md → "AI-SDD Mandates (M1-M7)" for BLOCKING criteria. When syncing docs after code changes, update the logical-ID mappings (FR-/BR-/OP-/TC-) FIRST, then the prose. The [Source: namespace/service/id] abstract-anchor evidence is re-resolved ONLY if the logical artifact was renamed/split — a file move or stack change does NOT change the anchor (physical coords live only in the provenance sidecar) — and the logical-ID spine stays stable across the change — never drop or renumber a logical ID just because the code moved. Keep all synced prose M1/M2-clean.

Be skeptical. Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80%.

Mandatory Task Creation (ZERO TOLERANCE)

[BLOCKING] Create ALL 8 tasks via task tracking BEFORE touching any file. NEVER consolidate, rename, omit. Conditional tasks skipped: mark completed immediately with reason — NEVER silently omit.

| # | Task Subject | Conditional? | | --- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------- | | 1 | [docs-update] Phase 0 — Triage: collect git diff, categorize files, detect modules, check existing docs | No — always first | | 2 | [docs-update] Phase 1 — Project context sync: impact-map → PARALLEL verify of impacted docs/project-reference/** + docs/project-config.json sections + README/project docs | No — always, unless Step 0.3 declared a TRUE fast exit (empty impact map). Runs even when Phases 2-4 are all skipped | | 3 | [docs-update] Phase 2 — Invoke $spec: update business feature docs | Yes — service/frontend files changed AND module has existing feature docs | | 4 | [docs-update] Phase 2.5/2.6 — Refresh derived views via $spec-index and/or $tech-spec | Yes — Feature Spec changed and bucket maintains INDEX/ERD, OR technical tree is affected | | 5 | [docs-update] Phase 3 — Invoke $spec [mode=tests]: update/add §8 business test specifications | Yes — business-visible functionality added OR existing business-visible behavior changed | | 6 | [docs-update] Phase 4 — Invoke $spec [mode=sync]: sync §8 ↔ test code | Yes — Phase 3 changed §8 TCs | | 7 | [docs-update] Phase 5 — Write summary report to plans/reports/docs-update-{YYMMDD}-{HHMM}.md | No — always | | 8 | [docs-update] Final review — verify all impacted docs updated, no phases skipped without justification, AND run the Step 2.4 code↔spec sync-verify (AC/BR/TC drift) for every touched module | No — always |

Execution rules:

Mark each task in_progress when starting, completed when done — one active at a time
Multiple modules → add one subtask per module for Phase 2/3 invocations
Multiple impacted reference docs → add one subtask per doc (or per source-of-truth cluster) under Task 2, so each doc's verdict is tracked individually
NEVER batch-complete — each sub-skill invocation tracked individually
Phase 0 TRUE fast-exit (impact map empty) → mark tasks 2-8 completed with reason "Skipped — impact map empty"
Phase 0 PARTIAL exit (docs/config impacted but no business behavior — e.g. harness, CI, or manifest-only changes) → run Task 2, mark tasks 3-6 completed with reason "Skipped — no business behavior changed", still run tasks 7-8
NEVER execute a phase step until matching task status is in_progress
After each phase/skill call, write one-line evidence in task update (what ran, what changed, why skipped)
If task tracking/task updates unavailable, maintain equivalent 8-task plan tracker with same status transitions

Step-Skill Call Order (Do Not Reorder)

| Order | Task ID | Step / Phase | Skill Call | Tracking Rule | | ----- | ------- | ----------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------ | | 1 | 1 | Phase 0: Triage | Inline triage logic in this skill | Set Task 1 in_progress before diff scan; set completed after module + impact map recorded | | 2 | 2 | Phase 1: Project Context Sync | doc-impact-map.cjs + PARALLEL docs-manager sub-agents (one per impacted doc/cluster) + $scan --target=X or $project-config only on escalation | Set Task 2 in_progress before the impact map; completed only after EVERY routed doc and config section carries a verdict + evidence | | 3 | 3 | Phase 2: Business Feature Docs | $spec | Set Task 3 in_progress before invocation; completed after output review | | 4 | 4 | Phase 2.5/2.6: Derived View Refresh | $spec-index [mode=index] and/or $tech-spec [mode=generate | audit] | Set Task 4 in_progress before invocation; completed after derived outputs are refreshed or skipped with reason | | 5 | 5 | Phase 3: §8 Test Specs | $spec [mode=tests] | Set Task 5 in_progress before invocation; completed after TC review | | 6 | 6 | Phase 4: §8 ↔ Test Code Sync | $spec [mode=sync] | Set Task 6 in_progress before invocation; completed after sync validation | | 7 | 7 | Phase 5: Summary Report | Inline report write | Set Task 7 in_progress before report write; completed after file path confirmed | | 8 | 8 | Final Review | Inline verification gate | Set Task 8 in_progress before final audit; completed after all phases justified |

Enforcement: If a required step cannot run, STOP and ask user before adapting order. Never continue with untracked steps.

Phase 0: Triage — Detect Impacted Documentation

Step 0.1: Collect Changed Files

Run git diff --name-only HEAD (staged + unstaged changes)
No uncommitted changes → git diff --name-only HEAD~1 (last commit)
Still empty → git diff --name-only origin/develop...HEAD (branch changes)

Step 0.2: Categorize Changes

| Changed File Pattern | Impact Category | Phases to Run | | ------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------- | ------------- | | {backend-source-paths}/** from docs/project-config.json | spec + spec [mode=tests] + project-docs | 1 + 2 + 3 + 4 | | {frontend-apps-dir}/**, {frontend-libs-dir}/{domain-lib}/** | spec + spec [mode=tests] + project-docs | 1 + 2 + 3 + 4 | | {legacy-frontend-dir}/**Client/** | spec + spec [mode=tests] + project-docs | 1 + 2 + 3 + 4 | | {configured-framework-source-paths}/** | project-docs only | 1 only | | docs/** (outside specRoots) | project-docs only | 1 only | | .claude/**, .agents/**, .codex/**, CLAUDE.md, AGENTS.md | harness inventory — skill/hook/agent/workflow counts, catalogs, module registry | 1 only | | Dependency manifests (package.json, *.csproj, pyproject.toml, lockfiles, …) | project-docs — tech stack, versions, run commands | 1 only | | Infra/CI/env (docker-compose*, Dockerfile, .github/workflows/**, *.tf, appsettings*, .env*) | project-docs — ports, deployment, env keys | 1 only | | {frontend-libs-dir}/{framework-core-lib}/**, {frontend-libs-dir}/{common-lib}/** | project-docs only | 1 only |

This table classifies BUSINESS-doc impact (which of Phases 2-4 run). It is deliberately coarse. The precise docs/project-reference/** + docs/project-config.json routing is produced by the impact map in Step 1.1 — read it there, never guess it here.

Step 0.3: Fast Exit Check — decided by the impact map, never by path intuition

Run the impact map NOW (same command as Step 1.1) and read fastExit from its output:

node .claude/scripts/doc-impact-map.cjs --text

| Map result | Route | | -------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------- | | fastExit: true — no impacted reference doc, no impacted config section, no unrouted file | Report "No documentation impacted by current changes." → mark tasks 2-8 completed with reason "Skipped — impact map empty" → exit early | | Impacted docs/config but NO business behavior changed (harness, CI, manifests, docs tree) | PARTIAL exit — run Phase 1 in full, mark Phases 2-4 completed with reason "Skipped — no business behavior changed", continue to Phase 5 | | Any business/service/frontend code changed | Full sequence | | unrouted non-empty | NOT a fast exit — classify each unrouted file by hand first (add it to the wave, or record why it carries no doc impact) |

[BLOCKING] A .claude/**-only (or tooling-only) diff is NOT a full fast exit. Harness edits change the skill/hook/agent/workflow inventories that CLAUDE.md, docs-index-reference.md, and project-structure-reference.mdderive by globbing.claude/: the counts and catalogs go stale with zero feature impact, and NO other gate in this skill catches them — Phases 2-4 only look at docs/specs/\*\*. — why: the cheapest way to ship stale docs is to classify the change as "tooling" and skip the only phase that would have noticed.

Step 0.4: Auto-Detect Affected Modules

Extract unique module names from changed paths. MUST ATTENTION dedup: unique() before passing to any sub-skill — backend + frontend same module = ONE entry. Prevents duplicate $spec invocations.

| Changed File Path Pattern | Detected Module | | --------------------------------------------------------------- | -------------------------------- | | {backend-module-path}/{Module}/** | {Module} | | {frontend-apps-dir}/{app-name}/** | {Module} (map app to module) | | {frontend-libs-dir}/{domain-lib}/{configured-feature-path}/** | {Module} (map feature to module) | | {legacy-frontend-dir}/{Module}Client/** | {Module} |

Build project-specific mapping from docs/project-config.json and project reference docs, not from hard-coded skill paths:

node -e "const cfg=require('./docs/project-config.json'); console.log(JSON.stringify({sourcePaths: cfg.codebaseHealth?.sourcePaths, contextGroups: cfg.contextGroups?.map(g => ({name:g.name,pathRegexes:g.pathRegexes})), specRoot: 'docs/specs/'}, null, 2))"
node -e "process.stdout.write('docs/specs/')"

Step 0.5: Check Existing Docs for Each Module

For each detected module:

Check the matching bucket directory exists under docs/specs/
Check that the bucket contains README.*.md Feature Specs, or use the project reference doc's feature-doc layout
Check the matching bucket directory under docs/specs/ exists using project reference docs
Record: hasFeatureSpec (§1–§7 present), hasTestSpecs (§8 present), hasDerivedIndex (bucket INDEX.md present)

Step 0.6: Declare the Doc-Update Wave

Detection is SEQ and comes FIRST — every assignment below is derived from the impacted-doc set, so nothing dispatches until Steps 0.1–0.5 have produced it. Once that set exists, updating N unrelated docs is embarrassingly parallel: one docs-manager sub-agent per doc or per doc cluster, all spawned in ONE message.

Declare before dispatch — Parallel plan: wave 1 = [docs-manager: {doc A}, docs-manager: {cluster B}, …] · SEQ = [Phase 0 triage, the Phase 2 → 2.5/2.6 → 3 → 4 spec chain, Phase 5 report] (reason).
STRICT one-writer-per-file. Every impacted doc path appears in EXACTLY ONE agent's brief, stated as that agent's owned file set. A doc owned by nobody is a silent miss; a doc owned by two agents is a lost-update race where the later write wins and the earlier finding vanishes.
[HAZARD] Two docs that embed the same canonical or derived data MUST go to the SAME agent — never split across the wave. Counts, catalogs, module maps, INDEX rows, ERD entities, and any table copied out of a source of truth have to be regenerated by ONE writer from ONE reading of that source. Split across two agents they diverge inside a single commit — and the divergence survives review because each doc is internally consistent and only the pair is wrong. Cluster by SOURCE OF TRUTH, not by directory.
Barrier before the spec chain. Phase 2 → 2.5/2.6 → 3 → 4 stays a FIXED SEQ chain: $spec output feeds the derived index, and §8 TCs feed [mode=sync]. Parallelism lives INSIDE a phase across independent modules — never across these phases.
Per-module fan-out is PAR only when the modules are disjoint. Two detected modules that map to ONE Feature Spec share a write target and stay a single task — the same reason Step 0.4 dedups the module list.
Every member returns a summary + Full report: path; YOU merge them into the Phase 5 report only after ALL members return, a skipped member counting as returned.

Phase 1: Project Context Sync — Reference Docs + project-config.json (PARALLEL, impact-scoped)

Why this phase exists. docs/project-reference/** and docs/project-config.json are injected into EVERY downstream AI context and route every skill in the framework. When code moves and they do not, the harness keeps teaching a codebase that no longer exists — and nothing else in this skill catches it, because Phases 2-4 only look at docs/specs/**. $scan-all + $project-config do repair them, but they re-derive every doc from zero, which is why they run every 60 days instead of every change. This phase does the same job at diff scope: same no-stale guarantee, small enough to afford after every change.

When to run: ALWAYS, unless Step 0.3 declared a TRUE fast exit. Run it even when every one of Phases 2-4 is skipped.

Scope discipline: verify ONLY what the impact map routes; escalate to a full $scan --target=X when a surgical patch cannot make the doc true again. NEVER regenerate all docs, and NEVER hand-author a full reference doc here — scan owns authoring, this phase owns verification and surgical repair.

Step 1.1: Build the Doc-Impact Map (SEQ — everything below derives from it)

node .claude/scripts/doc-impact-map.cjs --json     # machine-readable (drives the wave)
node .claude/scripts/doc-impact-map.cjs --text     # human-readable (goes in the report)
node .claude/scripts/doc-impact-map.cjs --base=origin/main   # branch-scope instead of working tree

The map routes each changed file to the docs and config sections it can rot, and returns per doc: doc, exists, lastScanned/ageDays, scanTarget (the full-rescan escalation), checks (which verifications apply), changedFiles/addedFiles/deletedFiles, and heuristicOnly. Routing is derived from docs/project-config.json (contextGroups, modules, testing, e2eTesting, styling, designSystem, specRoots) plus change-class rules — never from hardcoded project paths.

Handling the map's output — [BLOCKING] rules:

unrouted files are not proof of no impact — they are proof the router had no rule. Classify each by hand: add it to the wave, or record in the report why it carries no doc impact. NEVER let an unrouted file silently pass as fresh.
A heuristicOnly doc is a GUESS, not evidence. Verify it like any other, and downgrade to "not impacted" only with a stated reason.
Any doc whose exists: false is a MISSING doc, not a fresh one → route to $scan --target=<scanTarget> (or $docs-init when the whole set is absent).
If the script is unavailable (older checkout, non-Node host), derive the same map by hand from docs/project-config.json — match changed paths against contextGroups[].pathRegexes → guideDoc/patternsDoc/stylingDoc/designSystemDoc, modules[].pathRegex → project-structure + modules, test/e2e/styling paths → their docs, manifests → tech stack, infra/CI → ports & deployment, .claude/** → inventory counts. Record that the map was manual.

Step 1.2: Declare the Verify Wave (PAR — one message, all members)

Parallel plan: wave 1 = [docs-manager: {doc A}, docs-manager: {cluster B}, docs-manager: project-config.json, …] · SEQ = [Step 1.1 impact map, the Phase 2 → 2.5/2.6 → 3 → 4 spec chain, Phase 5 report] (reason)

Wave-construction rules — the Step 0.6 hazards apply verbatim, plus:

STRICT one-writer-per-file. docs/project-config.json has exactly ONE owning agent in the wave, always. Two agents merging JSON into the same file is a guaranteed lost update.
Cluster by SOURCE OF TRUTH, not by directory. README.md + project-structure-reference.md both restate the module/directory map, and CLAUDE.md + docs-index-reference.md + project-structure-reference.md all embed .claude/-derived counts — each such set goes to ONE agent so the numbers cannot diverge inside a single commit.
Every routed doc appears in exactly one brief. A doc owned by nobody is a silent miss.
Every member returns its verdict table + Full report: path; merge only after ALL members return.

Step 1.3: Per-Doc Verify Contract (what each wave member actually does)

Verify FIRST, patch NARROW. Run only the checks the map listed for that doc:

| Check | Question it answers | How to answer it | On failure | | ------------------- | ----------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | claims | Do the doc's cited paths/examples still exist? | node .claude/scripts/doc-impact-map.cjs claims <doc> (missing = dead, ambiguous = short-form that resolves by suffix), then grep each cited symbol at its cited file | Repoint or delete a dead citation, repo-root an ambiguous one (a citation is evidence — never leave a dead one). The reference-doc-freshness test suite fails the build on any dead citation | | coverage | Does every ADDED artifact of this doc's kind appear in it? | Diff the map's addedFiles against the doc's inventory/examples | Add the missing row/example with file:line | | counts | Do numeric claims match ground truth? | Re-derive by glob/grep (skills, hooks, agents, workflows, services, docs, tests) | Update the number — and the marker region if the count is generated | | conventions | Did the diff introduce a pattern the doc does not describe, or violate one it does? | Read the diff against the doc's rules | New pattern → document it. Violation → report it, do NOT document it as a convention | | commands | Do documented run/test commands still work? | Compare against manifests/scripts (package.json, test config, integrationTestVerify) | Patch the command | | versions | Do stated tech/framework versions match the manifests? | Read the manifest — never infer | Patch the version | | ports | Do documented ports/endpoints match infra config? | Read compose/k8s/appsettings — never infer | Patch the port | | links / catalog | Do cross-links and catalog rows resolve? | Existence-check each target | Fix or remove the row |

Verdict per doc (exactly one, evidence required):

| Verdict | Meaning | Required evidence | | ----------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------- | | FRESH | Every applicable check ran and passed; no edit needed | Which checks ran + what was compared | | PATCHED | Surgical edit applied, then $prompt-enhance <doc> run to keep it concise | Sections touched + file:line evidence for each new claim + prompt-enhance run confirmation (or stated reason skipped) | | RESCAN REQUIRED | Beyond surgical repair — a new subsystem/pattern family appeared, most of the impacted section's examples are dead, or the doc's structure no longer fits the code | The scanTarget to run ($scan --target=X), and whether it ran in this session or is queued | | UNVERIFIED | Could not be checked (missing tooling, blocked read, budget) | Why, and what must run next |

[BLOCKING] Never fabricate freshness. "Looks fine", "probably unchanged", and "the diff was small" are not check results. A doc that was not verified is UNVERIFIED, never FRESH — a false FRESH is worse than no check, because it retires the suspicion that would have caught the drift later.

[BLOCKING] Every PATCHED docs/project-reference/**doc MUST run$prompt-enhance <doc> (default --op=enhance) before the verdict is recorded. These docs are injected into every downstream AI context — a surgical edit that adds correct prose without re-compressing still leaves the doc bloated. $prompt-enhancekeeps content as concise as possible while staying valuable enough for AI (caveman compression + attention anchoring), closing the same gap on the narrow patch path that$scan --target=X's own mandatory final step (scan/SKILL.md Final Step) already closes on a full rescan — so a RESCAN REQUIRED doc that escalates to $scan gets it for free and needs no separate call here. Skip ONLY for a single stamp/date/count-only edit, and record the skip reason.

Step 1.4: project-config.json Drift Check (single writer, schema-validated)

Verify ONLY the sections the map flagged. For each:

Re-derive from evidence — read the changed files, not the old config value.
Surgical merge — add/update entries; NEVER rename, remove, or restructure a top-level section (the $project-config Schema Protection Rules apply here unchanged).
Prove every touched pathRegex/path still matches something real — a regex that matches zero files is stale config that silently disables every downstream router that depends on it, and no schema check catches it:

node -e "const c=require('./docs/project-config.json');const {execSync}=require('child_process');const files=execSync('git ls-files',{encoding:'utf8'}).split('\n').filter(Boolean).map(f=>'/'+f);for(const m of c.modules||[]){const re=new RegExp(m.pathRegex,'i');const n=files.filter(f=>re.test(f)).length;console.log((n?'OK  ':'DEAD')+' modules.'+m.name+' -> '+n+' file(s)')}"

Validate the schema after the merge:

node -e "const {validateConfig}=require('./.claude/hooks/lib/project-config-schema.cjs');console.log(JSON.stringify(validateConfig(require('./docs/project-config.json')),null,2))"

Escalate, don't improvise — a NEW top-level section, a new module class, a new tech stack, or a failed validation means the change is a re-scan, not a patch → run $project-config (and report that it is required if it cannot run in this session).

Step 1.5: README & Project Docs (docs-manager)

Pass the Phase 0 diff context to a docs-manager sub-agent (agent_type="docs-manager") for the prose project docs in the same wave:

README.md — update if project scope or setup changed (keep under 300 lines)
docs/project-reference/project-structure-reference.md — update if service architecture or cross-service patterns changed (same agent as README: shared source of truth)

Standalone invocation (not a workflow step) may first spawn 2-4 scout-external (preferred) or scout agents to locate affected areas; as a workflow step, skip scouting and use the Phase 0 diff context.

Exclusions unchanged: this agent NEVER owns docs/specs/**, test specs, spec-index/ERD, or derived technical views.

Step 1.6: Stamp Discipline (BLOCKING)

| What ran | Stamp to write | | ----------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | Full $scan --target=X |  (owned by scan, top of doc) | | Impact-scoped verify/patch on a doc that carries a Last scanned stamp |  on the line immediately AFTER that stamp | | Impact-scoped verify/patch on a doc with no Last scanned stamp — CLAUDE.md, .claude/docs/**, and any other AI-facing instruction file | Write NO stamp. Record the doc's verdict in the Step 1.7 output only. — why: those files read as live instruction, and .claude/agents/docs-manager.md:35 forbids provenance metadata in them; the no-meta-log rule wins over the stamp rule, it is not overridden by it. |

[BLOCKING] An impact-scoped pass MUST NOT touch Last scanned. That stamp is the input to the 60-day full-rescan gate (getStaleReferenceDocs → refreshScanStaleFlag, .claude/hooks/lib/session-init-helpers.cjs:769-810). Moving it from a partial check would buy speed by switching off the net that catches whole-doc rot — the two mechanisms are complementary, not interchangeable.

After any doc in this run WAS fully rescanned, re-evaluate the gate:

node -e "require('./.claude/hooks/lib/session-init-helpers.cjs').refreshScanStaleFlag()"

Step 1.7: Phase 1 Output

Emit the freshness table into the Phase 5 report — one row per routed doc plus one per config section, each with verdict, checks run, and evidence. Docs with RESCAN REQUIRED or UNVERIFIED are carried into Recommendations so the next session inherits the debt explicitly instead of silently.

Phase 2: Business Feature Documentation — Invoke `$spec`

When to run: Triage detected modules with hasFeatureDocs = true AND service/frontend files changed.

When to skip: No service/frontend feature files changed. Report: "No business feature docs impacted."

Step 2.1: Determine Create vs Update

| Scenario | Action | | ------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Module has existing feature docs | Invoke $spec — auto-detect triggers update flow | | Module has NO feature docs AND change adds/changes a feature (new endpoint, command/query, entity, business rule, user-facing behavior) | BLOCK — Report: "Module {Module} has NO Feature Spec but this change introduces feature behavior. Create the tech-free 8-section Feature Spec FIRST via $spec, then re-run docs-update." Do NOT skip. This is the doc-first gate. | | Module has NO feature docs AND change is tooling/style/config-only (no behavioral impact) | Skip with reason "No feature behavior changed — no Feature Spec required." (matches Phase 0 fast-exit at :113-120). | | User explicitly asked for full doc creation | Invoke $spec with explicit module name |

Step 2.2: Invoke `$spec`

$spec Update feature docs for modules: {detected modules}.
Changed files: {list from triage}.
Impacted sections based on change types: {section impact from triage}.
Mode: update (existing docs only, do not create from scratch).

What $spec handles (DO NOT duplicate here):

8-section tech-free structure enforcement
Diff analysis → section impact mapping
Codebase analysis (entities, commands, queries, controllers)
Update impacted sections with evidence
Bucket INDEX.md catalog row update
3-pass verification (evidence audit, domain model, cross-reference)
Tech-free principles (no implementation details in §1–§7; evidence carriers in §8 + [Source:] only)

Step 2.3: Review `$spec` Output

Updated sections align with triage's section impact mapping
No sections missed that triage flagged as impacted
Gaps found → re-invoke $spec for missed sections

Step 2.4: Code↔Spec Sync-Verify (final pass — runs because docs-update is last in every sequence)

Purpose: docs-update already runs LAST in feature/bugfix/big-feature, so this is the workflow's final gate. It is the order-time partner of the Phase 4 commit hook (this step guides; the hook enforces). Verify the SHIPPED code actually matches the mapped tech-free 8-section Feature Spec before the workflow completes.

For each module touched in this run, diff the changed code against its Feature Spec and check three sets:

| Spec set (Feature Spec section) | Sync check against changed code | On drift | | -------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------ | -------- | | §3 Acceptance Criteria (AC-{FC}-NN) | Every changed user-facing behavior maps to an AC; new behavior with no AC = missing AC. | Report drift; re-invoke $spec to add the AC. | | §4 Business Rules (BR-{FC}-NNN, [HARD]/[SOFT]) | Each changed validation/invariant matches a BR; a [HARD] rule whose code path was removed/weakened = regression. | BLOCK — surface as a code-vs-spec contradiction for the author to resolve. | | §8 Test Specifications (TC-{FC}-NNN + CoveredBy:) | Each new/changed business-visible behavior has a TC; each Tested TC's CoveredBy: {File}::{Method} or approved coverage carrier still resolves. Legacy IntegrationTest: is migration input only. | Report; route to $spec [mode=sync]. | | Derived technical views (specRoots.technical.path) | Technical-only coverage or component topology changes may require a regenerated/audited derived view. The view is generated from code/tests and is never hand-authored. | Report; route to $tech-spec [mode=generate | audit]. |

Output: a short sync-verify table (module · AC drift · BR drift/contradiction · TC drift) appended to the docs-update report. Clean = no drift across all three. A [HARD]-BR contradiction blocks workflow completion until resolved or explicitly accepted by the owner.

Scope: business code↔spec drift only. Technical contracts (API routes/DTOs, bus/job mechanics) are code-canonical and intentionally NOT re-verified against prose. No new sequence step and no verify-sync mode is added — this responsibility lives inside docs-update's existing final pass.

Phase 2.5: Derived Index / ERD Refresh (OPTIONAL — spec-index)

[SINGLE-HOME] There is no separate "engineering spec bundle". The canonical artifact is the 8-section Feature Spec updated in Phase 2. spec-index is repurposed to regenerate only the DERIVED bucket INDEX.md / cross-capability ERD from those Feature Specs — it never re-extracts an A-E tree. Run this phase only if the bucket maintains a derived index/ERD that the Phase 2 change made stale.

When to run: Phase 2 changed one or more Feature Specs AND the bucket maintains a derived INDEX.md / ERD aid that now lags.

When to skip:

Only docs/, .claude/, or config files changed
No Feature Spec under docs/specs/{Bucket}/ was touched
Phase 2 was skipped (no feature impact)
The bucket maintains no derived index/ERD, OR project-config.json contains "spec_discovery_update": false
spec already refreshed INDEX.md in Phase 2 (no separate refresh needed)

Step 2.5.1: Resolve the Bucket

Map the changed services to an App Bucket using the canonical table in docs/project-reference/spec-system-reference.md → App Bucket Mapping.
Confirm docs/specs/{Bucket}/ holds the updated Feature Spec(s).

Step 2.5.2: Invoke spec-index (Derived Index Mode)

$spec-index mode=index bucket={Bucket} artifacts=INDEX[,ERD]
Source: the canonical Feature Specs in docs/specs/{Bucket}/.
Output: regenerated DERIVED docs/specs/{Bucket}/INDEX.md (+ {Bucket}.erd.md if maintained), each carrying the DERIVED banner.

Step 2.5.3: Verify Refresh Complete

Confirm INDEX.md rows match the current set of Feature Specs (no dangling links, no missing capabilities).
Confirm the DERIVED banner + regenerate date are present.
Report: "Derived index refreshed: {Bucket} — {N} capabilities catalogued".

Separation of concerns: docs-update orchestrates — passes the bucket scope to spec-index. NEVER hand-edits the derived index, and NEVER recreates M##/A-E artifacts (retired).

Phase 3: Test Specifications — Invoke `$spec [mode=tests]`

When to run: New business-visible functionality added OR existing business-visible behavior changed. Technical-only changes with no changed user/QC-visible outcome produce no business Section 8 edits; route any technical coverage need to tests and $tech-spec for the derived technical view.

When to skip: Changes purely cosmetic (styling, comments, docs-only) with no behavioral impact.

Step 3.1: Determine TC Mode

| Context | TC Mode | | -------------------------------------- | ------------------------ | | New feature code, no existing TCs | implement-first | | PBI/story exists, code not yet written | TDD-first | | Existing TCs + code changes / bugfix | update | | User says "sync test specs" | sync | | Tests exist with annotations, no docs | from-integration-tests |

PBI/idea artifact route: when changed artifacts match configured PBI/idea artifact roots from docs/project-config.json or project reference docs, docs-update performs detection/delegation only. It may identify affected module, feature doc, and TC scope, then route to $spec, $spec [mode=tests], or $spec [mode=sync]. It must not generate TC content directly from PBI/idea artifacts or edit Section 8 itself. If artifact roots are not configured, ask the user to initialize project config/reference docs before assuming a path.

Step 3.2: Invoke `$spec [mode=tests]`

$spec [mode=tests] Mode: {detected mode}.
Modules: {detected modules}.
Changed files: {list from triage}.
Business-visible functionality detected: {new or changed user/QC-visible outcomes from diff analysis}.

What $spec [mode=tests] handles (DO NOT duplicate here):

5 modes: TDD-first, implement-first, update, sync, from-integration-tests
TC-{FEATURE}-{NNN} format with decade-based numbering
Interactive TC review (ask the user directly)
Cross-cutting categories: authorization, seed data, performance, data migration
Phase-mapped coverage (plan phases → TCs)
Graph context analysis for cross-service impact
Evidence verification per TC
Write to feature doc Section 8 (canonical business TC registry)

Step 3.3: Review `$spec [mode=tests]` Output

New TCs cover all new business-visible functionality from triage
TC IDs don't collide with existing ones
Evidence fields populated (not template placeholders)

Phase 4: Test Spec ↔ Test Code Sync — Invoke `$spec [mode=sync]`

When to run: Phase 3 produced new/updated TCs in §8 of a Feature Spec.

When to skip: No §8 test-spec changes.

Step 4.1: Invoke `$spec [mode=sync]`

$spec [mode=sync] Sync test specs for capabilities: {detected features}.
Direction: forward (Feature Spec §8 Test Specifications → executing test code).
Updated TCs from Phase 3: {list of new/changed TC IDs}.

What $spec [mode=sync] handles (DO NOT duplicate here):

Forward/reverse sync: §8 Test Specifications ↔ executing test code
2-way comparison: Feature Spec §8 vs test code (code is the technical source of truth)
Test cross-reference (configured test-spec annotation key TestSpec across all executing test tiers and the per-TC CoveredBy: field; legacy IntegrationTest: is migration input only)

The retired QA dashboards (docs/specs/README.md, docs/specs/PRIORITY-INDEX.md) and the hand-maintained A-E/M## engineering tree no longer exist — §8 is the canonical business TC registry. Derived aids are the bucket INDEX.md count (Phase 2.5) and the regenerable technical view under specRoots.technical.path (Phase 2.6).

Step 4.2: Review Sync Results

All new TCs from Phase 3 are reflected in test code (or flagged Untested with rationale).
No orphaned TCs (referenced by test code's TestSpec annotation but absent from §8).

Section Ownership Reference

Which skill owns which doc sections — docs-update delegates only, NEVER writes directly:

| Section | Owner Skill | docs-update Role | | --------------------------------------- | ------------------------ | ------------------------------------------------------------------------------------------------------------------------------------- | | docs/project-reference/** (authoring) | $scan --target=X | Verify impact-scoped + surgical patch; escalate to the scan when a patch cannot make it true — NEVER hand-author a full reference doc | | docs/project-config.json (structure) | $project-config | Verify + surgical merge of impacted sections with schema validation; escalate whole-section/new-tech changes | | §1–§7 (Feature Spec, tech-free) | $spec | Pass triage context; review output | | §8 (Test Specifications) | $spec [mode=tests] | Pass TC mode + changed files; NEVER write TCs here | | §8 ↔ test code sync | $spec [mode=sync] | Pass capability list + direction; NEVER edit directly | | Derived bucket INDEX.md / ERD | $spec-index (optional) | Pass bucket scope; NEVER hand-edit the derived index | | Derived technical spec view | $tech-spec (optional) | Pass service/component scope; NEVER hand-edit the derived technical file |

Phase 5: Summary Report

ALWAYS write full report to plans/reports/docs-update-{YYMMDD}-{HHMM}.md:

### Documentation Update Summary

**Triage:** {N} files changed → {categories detected}
**Modules detected:** {module list}
**Generated mirror sync:** {Completed / N/A / Required before close}

**Phase 1 — Project Context Sync (impact-scoped freshness):**

Impact map: {N} changed files → {D} docs, {S} config sections, {U} unrouted ({map source: working tree / last commit / branch})

| Reference doc | Verdict                                        | Checks run                    | Evidence / action                                         |
| ------------- | ---------------------------------------------- | ----------------------------- | --------------------------------------------------------- |
| {doc}         | FRESH / PATCHED / RESCAN REQUIRED / UNVERIFIED | {claims, coverage, counts, …} | {what was compared; sections patched; scan target queued} |

| project-config.json section | Verdict                                        | Evidence / action                                                   |
| --------------------------- | ---------------------------------------------- | ------------------------------------------------------------------- |
| {section}                   | FRESH / PATCHED / RESCAN REQUIRED / UNVERIFIED | {re-derived from …; schema validation result; dead pathRegex found} |

- Unrouted files classified: {file → why no doc impact}
- README / project docs: {Updated/Skipped}: {reason}
- Stamps: {docs given `Last verified`} · {docs fully rescanned and given `Last scanned`} · staleness flag refreshed: {yes/no}

**Phase 2 — Feature Specs ($spec):**

- {Capability X}: {Updated §1–§7 / No existing Feature Spec / Not impacted}
- {Capability Y}: {Updated §4 Business Rules, §5 Domain Model / Skipped: no Feature Spec}

**Phase 2.5 — Derived Index Refresh ($spec-index, optional):**

- {Refreshed {Bucket} INDEX.md ({N} capabilities) / Skipped: no derived index maintained / Skipped: spec_discovery_update=false}

**Phase 3 — Test Specifications §8 ($spec [mode=tests]):**

- Mode: {mode used}
- New TCs: {list of TC IDs added}
- Updated TCs: {list of TC IDs modified}
- Skipped: {reason if skipped}

**Phase 4 — Test Spec ↔ Test Code Sync ($spec [mode=sync]):**

- {Synced N TCs to test code / Skipped: no §8 changes}
- Discrepancies: {§8-vs-test-code comparison issues}

**Recommendations:**

- {New docs that should be created}
- {Stale docs flagged but not auto-fixed}
- {TCs flagged as Untested}

Decision Matrix: When to Use docs-update vs Direct Skill

| Scenario | Use docs-update? | Use skill directly? | | ------------------------------------------------------------------------ | -------------------------------------- | -------------------------------------------------------------------------------------- | | Post-implementation doc sync (any code change) | Yes — full orchestration | — | | Keep project-reference docs + project-config.json fresh after a change | Yes — Phase 1 impact-scoped verify | $scan --target=X or $project-config when a single doc/section needs a full rebuild | | Refresh every reference doc regardless of the diff | No | $scan-all (+ $project-config) | | Create new feature docs from scratch | No | $spec | | Generate TCs for specific PBI (TDD-first) | No | $spec [mode=tests] | | Route PBI/idea artifact changes | Yes — detection/delegation | $spec + $spec [mode=tests] owner skills | | Sync dashboard only (no code changes) | No | $spec [mode=sync] | | Workflow step after $plan-execute or $fix | Yes — full orchestration | — | | User asks "update docs after my changes" | Yes — full orchestration | — |

Additional Requests

Pass caller context via $ARGUMENTS to skip redundant triage or narrow scope:

| Key | Example | Effect | | --------------- | ---------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | modules | modules=ModuleA,ModuleB | Skip auto-detect; use provided list | | changed_files | changed_files=<configured-source-path>/ModuleA/... | Skip git diff; use provided file list | | phases | phases=2,3 | Run only specified phases | | mode | mode=update | Override spec mode detection | | tc_mode | tc_mode=implement-first | Override spec [mode=tests] mode detection | | skip_phases | skip_phases=1,2.5 | Skip specific phases | | freshness | freshness=impact (default) / full / off | impact = Phase 1 as specified; full = escalate every routed doc to its $scan --target=X (and $project-config); off = skip Phase 1 — allowed ONLY on explicit user instruction, and the report MUST record every routed doc as UNVERIFIED | | base | base=origin/main | Scope the impact map to a branch diff instead of the working tree |

<additional_requests> $ARGUMENTS </additional_requests>

Escalation: When docs-update Is Not Enough

| Situation | What to do instead | | -------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ | | Feature Spec missing but capability exists | Run $spec [mode=init] to author the 8-section Feature Spec, then docs-update | | Derived bucket INDEX.md/ERD missing | Run $spec-index mode=index bucket={Bucket} to (re)generate it | | Integration tests don't match TCs | Run $integration-test-review to diagnose, then $integration-test to fix | | Bug caused by wrong spec | Run $spec [mode=update] (fix the canonical spec) BEFORE docs-update; optionally $spec-index mode=index to re-derive the bucket index | | One reference doc is wrong beyond a surgical patch | Run its $scan --target=<key> (the map's scanTarget), then re-run Phase 1 to confirm | | Most reference docs are stale, or the 60-day gate fired | Run $scan-all — impact scope cannot repair rot that predates the diff | | A reference doc does not exist at all | Run $docs-init (whole set missing) or $scan --target=<key> (single doc) | | project-config.json needs a new section, module class, or tech stack, or fails schema validation | Run $project-config — that is a re-scan, not a merge | | Suspect long-standing rot that no current diff touches | Run $scan-codebase-health (count-drift, dead config references, broken cross-links) |

[BLOCKING] Create ALL 8 tasks via task tracking BEFORE any action — see Mandatory Task Creation table. NEVER skip, batch-complete, or mark done without invoking sub-skill. [BLOCKING] Follow fixed step-skill order: Phase 0 -> Phase 1 -> Phase 2 -> Phase 2.5/2.6 -> Phase 3 -> Phase 4 -> Phase 5 -> Final review. NEVER reorder, merge, or skip without explicit user approval. [BLOCKING] Per-step task lock: BEFORE each step, mark task in_progress; AFTER each step, mark task completed with evidence or explicit skip reason. [BLOCKING] If Task tool unavailable, create equivalent 8-step plan tracker and keep statuses synced for every step.

Critical Purpose: Single orchestrator for ALL documentation sync after code changes. Triages impact, delegates to specialized skills.

Evidence Gate: [BLOCKING] — every claim requires file:line proof or traced evidence, confidence >80% to act.

Critical Thinking Mindset — Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80% to act. Anti-hallucination: Never present guess as fact — cite sources for every claim, admit uncertainty freely, self-check output for errors, cross-reference independently, stay skeptical of own confidence — certainty without evidence root of all hallucination.

Sub-Agent Return Contract — When this skill spawns a sub-agent, the sub-agent MUST return ONLY this structure. Main agent reads only this summary — NEVER requests full sub-agent output inline.
## Sub-Agent Result: [skill-name]

Status: ✅ PASS | ⚠️ PARTIAL | ❌ FAIL
Confidence: [0-100]%

### Findings (Critical/High only — max 10 bullets)

- [severity] [file:line] [finding]

### Actions Taken

- [file changed] [what changed]

### Blockers (if any)

- [blocker description]

Full report: plans/reports/[skill-name]-[date]-[slug].md
Main agent reads Full report file ONLY when: (a) resolving a specific blocker, or (b) building a fix plan. Sub-agent writes full report incrementally (per SYNC:incremental-persistence) — not held in memory.

Context budget — the return payload is a SUMMARY, not a transcript: ≤10 finding bullets, no raw file contents / full diffs / verbatim logs inline, no re-pasted source. Everything beyond the summary lives in the Full report on disk. A sub-agent that would exceed the summary shape MUST write the detail to its report and return only the pointer — the orchestrator's context is the scarce resource the whole map-reduce protects.

Cross-Service Check — Microservices/event-driven: MANDATORY before concluding investigation, plan, spec, or feature doc. Missing downstream consumer = silent regression.

| Boundary | Grep terms | | ------------------- | ------------------------------------------------------------------------------- | | Event producers | Publish, Dispatch, Send, emit, EventBus, outbox, IntegrationEvent | | Event consumers | Consumer, EventHandler, Subscribe, @EventListener, inbox | | Sagas/orchestration | Saga, ProcessManager, Choreography, Workflow, Orchestrator | | Sync service calls | HTTP/gRPC calls to/from other services | | Shared contracts | OpenAPI spec, proto, shared DTO — flag breaking changes | | Data ownership | Other service reads/writes same table/collection → Shared-DB anti-pattern |

Per touchpoint: owner service · message name · consumers · risk (NONE / ADDITIVE / BREAKING).

BLOCKED until: Producers scanned · Consumers scanned · Sagas checked · Contracts reviewed · Breaking-change risk flagged

AI Mistake Prevention — Failure modes to avoid on every task:

Re-read files after context changes. Context compaction, resume, or long-running work can make memory stale; verify current files before acting. Verify generated content against source evidence. AI hallucinates APIs, names, claims, and document facts. Check the relevant source before documenting or referencing. Check downstream references before deleting or renaming. Removing an artifact can stale docs, generated mirrors, configs, and callers; map references first. Trace the full impact chain after edits. Changing a definition can miss derived outputs and consumers. Follow the affected chain before declaring done. Verify ALL affected outputs, not just the first. One green check is not all green checks; validate every output surface the change can affect. Assume existing values are intentional — ask WHY before changing OR flagging one as a defect. Before changing or reporting a constant, limit, flag, cutoff, wording, or pattern, read nearby context and history, the CALLER's ordering, and 2+ sibling call sites of the same convention. A doc stating WHAT without WHY is missing rationale, not proof of a missing guard. Surface ambiguity before acting — don't pick silently. Multiple valid interpretations require an explicit question or stated assumption with risk. Assert the outcome your system owns, not the intermediate state your infrastructure owns. When verifying async work, assert the final business state — never the delivery/retry bookkeeping held in shared infrastructure that any co-running process can write. Such a check passes when run alone and flakes the moment anything else shares that infrastructure. Keep shared guidance role-relevant. Universal guidance must help every receiving skill or agent; code-specific obligations belong only in code-specific protocols.

Nested Task Expansion Contract — For workflow-step invocation, the [Workflow] ... row is only a parent container; the child skill still creates visible phase tasks.

Call the current task list first. If a matching active parent workflow row exists, set nested=true and record parentTaskId; otherwise run standalone.

Create one task per declared phase before phase work. When nested, prefix subjects [N.M] $skill-name — phase.

When nested, link the parent with TaskUpdate(parentTaskId, addBlockedBy: [childIds]).

Orchestrators must pre-expand a child skill's phase list and link the workflow row before invoking that child skill or sub-agent.

Mark exactly one child in_progress before work and completed immediately after evidence is written.

Complete the parent only after all child tasks are completed or explicitly cancelled with reason.

Blocked until: the current task list done, child phases created, parent linked when nested, first child marked in_progress.

Project Reference Docs Gate — Run after task-tracking bootstrap and before target/source file reads, grep, edits, or analysis. Project docs override generic framework assumptions.

Identify scope: file types, domain area, and operation.

Read docs/project-config.json first — the project's machine-readable map. It is the single source of truth for THIS repo (modules/paths, framework + search keywords, test/E2E/integration run-commands, design system, architecture rules, workflow patterns); ground exact paths, run-commands, and conventions on it before investigating, planning, or coding — never assume framework defaults (CLAUDE.md + reference docs are derived from it). If it — or the docs index, lessons.md, CLAUDE.md, AGENTS.md, or any required reference doc — is missing or stale, auto-run $project-init or the narrow route ($project-config, $docs-init, $scan-all, $scan --target=<key>, $claude-md-init) first; if Codex mirrors or AGENTS.md are stale, ask the user to run $sync-codex (never auto-run it).

Required docs by trigger: always docs/project-reference/lessons.md; doc lookup docs-index-reference.md; review code-review-rules.md; backend/CQRS/API backend-patterns-reference.md; domain/entity domain-entities-reference.md; frontend/UI frontend-patterns-reference.md; styles/design scss-styling-guide.md + design-system/design-system-canonical.md; integration tests integration-test-reference.md; E2E e2e-test-reference.md; feature docs/specs feature-spec-reference.md + spec-system-reference.md + spec-principles.md; behavior/public-contract/spec-test-code sync workflow-spec-test-code-cycle-reference.md; derived spec index/ERD/reimplementation guides spec-system-reference.md + source Feature Specs under docs/specs/; architecture/new area project-structure-reference.md.

Read every required doc, then before target work state: Reference docs read: ... | Not applicable: ....

Ready when: scope evaluated, docs/project-config.json consulted, required docs checked/read or setup route completed, lessons.md confirmed, citation emitted.

Task Tracking & External Report Persistence — Bootstrap this before execution; then run project-reference doc prefetch before target/source work.

Create a small task breakdown before target file reads, grep, edits, or analysis. On context loss, inspect the current task list first.

Mark one task in_progress before work and completed immediately after evidence; never batch transitions.

For plan/review work, create plans/reports/{skill}-{YYMMDD}-{HHmm}-{slug}.md before first finding.

Append findings after each file/section/decision and synthesize from the report file at the end.

Final output cites Full report: plans/reports/{filename}.

Blocked until: task breakdown exists, report path declared for plan/review work, first finding persisted before the next finding.

MUST ATTENTION apply critical + sequential thinking — every claim needs appropriate traced evidence (file:line for repo/code claims; source URL or artifact section for research, product, content, and docs claims); confidence >80% to act, <60% DO NOT recommend. Anti-hallucination: never present guess as fact, admit uncertainty freely, cross-reference independently, stay skeptical of own confidence.

MUST ATTENTION apply AI mistake prevention — verify generated content against evidence, trace downstream references before deleting or renaming, verify all affected outputs, re-read files after context loss, and surface ambiguity before acting.

MANDATORY Bootstrap task tracking before target work; transition one task at a time.
MANDATORY Persist plan/review findings to plans/reports/ incrementally and synthesize from disk.

MANDATORY Before investigating, planning, or coding, read docs/project-config.json (the project map: modules/paths, run-commands, conventions, architecture/workflow rules) + the required project-reference docs, and cite Reference docs read: ....
MANDATORY Always include lessons.md; project config + conventions override generic framework defaults.
MANDATORY If project config, root instruction files, or any required reference doc is missing or stale, auto-run $project-init or the narrow lower-level route before ordinary project-specific work.

MANDATORY Parent workflow rows do not replace child phase tracking; expand phases and link the parent when nested.
MANDATORY Orchestrators pre-expand child skill phases before invocation; use [N.M] $skill-name — phase prefixes and one-in_progress discipline.

Prompt-Enhance Closing Anchors

IMPORTANT MUST ATTENTION follow declared step order for this skill; NEVER skip, reorder, or merge steps without explicit user approval IMPORTANT MUST ATTENTION for every step/sub-skill call: set in_progress before execution, set completed after execution IMPORTANT MUST ATTENTION every skipped step MUST include explicit reason; every completed step MUST include concise evidence IMPORTANT MUST ATTENTION if Task tools unavailable, maintain an equivalent step-by-step plan tracker with synchronized statuses

Parallel Sub-Agent Dispatch — Plan parallelism the moment a task breakdown exists, BEFORE executing it — running provably independent tasks sequentially wastes wall-clock. Applies to every multi-step job: workflow steps, planning, batch updates, investigation, research, scans, reviews, doc sync. Plan execution is metadata-gated, NEVER default-parallel — fan-out follows ONLY what the plan declares (PAR/SEQ tags + per-phase write set); an untagged plan runs sequentially — why: a derived write set cannot see cascade or generated writes.

Tag every task PAR or SEQ. PAR = inputs exclude every pending task's output AND write set disjoint from every other PAR. Else SEQ — MUST ATTENTION name the dependency forcing it.

Group PAR into waves. No edge between members. Two writers of one file NEVER share a wave. Read-only work (search, investigation, review, research) parallelizes freely.

Declare before dispatch: Parallel plan: wave 1 = [...] · wave 2 = [...] · SEQ = [...] (reason).

Spawn each wave in ONE message — every spawn_agent call in one response, NEVER dripped per turn. Route each task to its specialist (.claude/skills/shared/sub-agent-selection-guide.md); NEVER code-reviewer as catch-all.

Brief each sub-agent self-contained: goal · scope + owned files · reference docs · return contract (summary + Full report: path, per SYNC:subagent-return-contract) · incremental persistence to plans/reports/ (per SYNC:incremental-persistence).

Barrier per wave. Advance ONLY after EVERY member returns (a skipped conditional counts as returned). Merge, mark each task completed/skipped, THEN dispatch the next wave. Mutating steps wait for the barrier.

One level deep. A dispatched sub-agent executes its own brief; further fan-out stays the orchestrator's job unless that agent's .claude/agents/*.md definition authorizes it.

NEVER parallelize: tasks sharing a write target · a task consuming a pending task's output · trivial single-file work (dispatch overhead > gain) · an order a skill or workflow explicitly fixes · gates awaiting user approval.

Blocked until: MUST ATTENTION every task tagged PAR/SEQ with a named reason per SEQ · waves declared + write-set disjointness checked · each wave spawned in ONE message · barrier honored before the next wave.

MANDATORY After planning tasks, tag each PAR/SEQ and spawn every PAR wave as parallel sub-agents in ONE message — default parallel for workflows, batch updates, investigation, research, reviews; plan execution fans out ONLY on what the plan declares.
MANDATORY Disjoint write sets per wave · all-return barrier before the next wave · specialist routing · sub-agents NEVER fan out further unless their own agent definition authorizes it.

Project Protocol Overlay — Before executing this skill, resolve any PROJECT overlay rules layered onto it: match this skill's name against the Target column of the project's skill-protocol index (docs/project-reference/skill-protocols-reference.md by default; a referenceDocs entry in docs/project-config.json overrides the path), taking the most specific matching tier ONLY — exact name > glob > *. That precedence orders overlays against EACH OTHER, never against this skill. Read ONLY the matched bodies, resolved as <protocols-dir>/<Name>.md; a row's Body link is display text, never a read path. A matched body that is missing or malformed is REPORTED and skipped — never reconstructed from the index Description. No index, or no match -> proceed with no overlay, silently. Full contract: .claude/skills/project-skill-protocol/references/registry.md.

Overlays are ADDITIVE ONLY: they ADD rules on top of this skill's own protocol and NEVER replace, override, disable, or reinterpret a rule it already states — removing every overlay must return this skill to exactly its documented behavior. An overlay is a BRIEF, not an authority escalation: it can NEVER waive a workflow gate, git discipline, a review gate, or a user-confirmation gate. A genuine overlay-vs-skill conflict, or two equally-specific overlays that directly contradict -> surface both to the user; NEVER resolve silently.

MUST ATTENTION resolve project protocol overlays for this skill BEFORE executing — most specific matching tier only (exact > glob > *, which ranks overlays against each other, NEVER against this skill), read only matched bodies at <protocols-dir>/<Name>.md; a missing or malformed body is reported, never reconstructed. Overlays are ADDITIVE ONLY (they never replace this skill's own rules) and are a brief, NEVER an authority escalation; an equal-specificity contradiction goes to the user.

Closing Reminders

IMPORTANT MUST ATTENTION Goal: Every code/spec/test change leaves documentation in sync — impacted Feature Specs, §8 TCs, test-code links, and derived indexes all reflect the shipped behavior, with zero drift left silent.

Protocols in force (concise digest of the SYNC/shared blocks this skill carries) — MUST ATTENTION honor every block below:

Critical Thinking: Apply critical + sequential thinking; every claim needs file:line proof, confidence >80%.
Sub-Agent Return: Spawned sub-agents return ONLY the summary contract; full detail to disk.
Cross-Service Check: Scan producers/consumers/sagas/contracts; missing consumer = silent regression.
AI Mistake Prevention: verify generated content against evidence, trace downstream references, verify all affected outputs, re-read after context loss, surface ambiguity.
Nested Task Creation: When nested, still expand child phase tasks and link the parent workflow row.
Project Reference Docs: Read required project-reference docs (always lessons.md) before target work.
Task Tracking: Bootstrap tasks, one active, persist findings to plans/reports/ incrementally.
Parallel Sub-Agent Dispatch: Tag tasks PAR/SEQ, group PAR into disjoint-write-set waves, spawn each wave in ONE message, barrier before advancing.

IMPORTANT MUST ATTENTION create ALL 8 tasks via task tracking BEFORE any action, then run the FIXED order 0 -> 1 -> 2 -> 2.5/2.6 -> 3 -> 4 -> 5 -> final review — NEVER reorder, merge, or skip without explicit user approval — why: phase order is the gate that catches drift; a skipped phase ships silent staleness IMPORTANT MUST ATTENTION docs-update is a ROUTER ONLY — delegate to $spec, $spec [mode=tests], $spec [mode=sync], $spec-index; NEVER write §8 content, edit Feature Spec / derived-index files, or duplicate sub-skill logic — why: dual authorship causes the two sources to diverge IMPORTANT MUST ATTENTION every skip is a DECISION with evidence — mark the task completed with a file:line-backed reason; NEVER silently omit a phase — why: an unjustified skip is indistinguishable from a missed update MUST ATTENTION Nested Task Expansion Contract — when invoked inside a workflow, STILL expand internal phases via task tracking with [N.M] $skill-name — phase prefix and TaskUpdate(parentTaskId, addBlockedBy: [childIds]) linkage — why: the workflow row is a container, not a substitute for phase tracking MUST ATTENTION for EVERY step: set task in_progress BEFORE execution, set completed AFTER execution with evidence or skip reason — never batch transitions, keep exactly one active MUST ATTENTION if task tooling unavailable, use an equivalent 8-step plan tracker and keep statuses synced per step MUST ATTENTION evidence gate — every claim, detected module, and impact mapping needs file:line / git-diff proof, confidence >80% to act, <60% DO NOT act; "Module unchanged" without proof is NOT a valid skip — why: speculation routes the wrong docs and misses real drift MUST ATTENTION search-existing-patterns BEFORE asserting a doc shape — read the bucket's existing Feature Spec / INDEX layout and project-reference docs; build the module map from docs/project-config.json, NEVER from hard-coded skill paths — why: local doc conventions override generic assumptions MUST ATTENTION evaluate fit before reusing a nearby pattern — a module with backend + frontend changes is ONE deduped entry, not two; verify the change actually alters behavior before routing to $spec — why: duplicate or behavior-free invocations waste passes and corrupt the audit MUST ATTENTION validate ambiguous routing decisions with the user by asking the user directly — surface the options, NEVER silently auto-decide which phases run MUST ATTENTION tech-agnostic output — when updating spec/specs/README/INDEX, introduce NO framework/product/language/pattern names in prose or headings; update logical IDs (FR-/BR-/OP-/TC-) FIRST, then prose; preserve the evidence-field exception — why: prose is the portable contract, evidence carriers hold the physical coords (spec-principles §3) MUST ATTENTION Step 2.4 final code↔spec sync-verify per touched module — a removed/weakened [HARD] BR is a code-vs-spec contradiction that BLOCKS completion until resolved or owner-accepted; AC drift re-invokes $spec, TC drift routes to $spec [mode=sync] MUST ATTENTION Phase 1 project context sync ALWAYS runs unless the impact map is empty — build the map (node .claude/scripts/doc-impact-map.cjs), verify the routed docs/project-reference/** docs and docs/project-config.json sections in a PARALLEL wave, give every routed doc a verdict (FRESH | PATCHED | RESCAN REQUIRED | UNVERIFIED), and NEVER move a Last scanned stamp from an impact-scoped pass — why: these docs feed every downstream AI context, an unchecked doc is UNVERIFIED not FRESH, and a moved stamp silently disables the 60-day full-rescan gate MUST ATTENTION Phase 0 triage ALWAYS runs first (git diff → categorize → dedup modules → record existing-doc state); Phase 2 $spec updates §1–§7 and BLOCKS doc-first when a changed module has feature behavior but no Feature Spec — why: skipping triage or the doc-first gate ships undocumented behavior MUST ATTENTION Phase 2.5 $spec-index [mode=index] OPTIONALLY refreshes the derived bucket INDEX/ERD from Feature Specs (never re-extracts an A-E tree); Phase 2.6 $tech-spec OPTIONALLY refreshes/audits the derived technical view; Phase 3 $spec [mode=tests] syncs §8 TCs; Phase 4 $spec [mode=sync] syncs §8 TCs ↔ executing test code (no QA dashboard exists) MUST ATTENTION for .claude skills/hooks/workflows/sync-tooling changes, flag generated-mirror sync status (npm run codex:sync completed or explicit N/A) — docs-update routes/reports this check, NEVER edits generated mirrors directly MUST ATTENTION ALWAYS write the Phase 5 summary report to plans/reports/docs-update-{YYMMDD}-{HHMM}.md and the final review task (#8) — the report is the audit trail, the review verifies all impacted docs updated with no unjustified skips

Anti-Rationalization:

| Evasion | Rebuttal | | ---------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------- | | "Only docs/config changed — skip all phases" | Run Phase 0 triage anyway — fast-exit is a DECISION, not an assumption | | "Only .claude/** changed — tooling, fast exit" | Harness edits rot glob-derived inventory counts and catalogs in CLAUDE.md / docs-index / project-structure. Phase 1 STILL runs; only Phases 2-4 are skipped | | "The reference docs looked fine" | "Looked fine" is not a check. Run the doc's routed checks or report it UNVERIFIED | | "I updated the doc, so I'll refresh Last scanned" | Only a full $scan --target=X may move that stamp — an impact-scoped patch writes Last verified, and only where a Last scanned stamp already exists | | "project-config.json is close enough" | Re-derive the flagged sections from evidence, prove every touched pathRegex still matches a real file, and run schema validation | | "The mapper returned unrouted files — nothing to do" | Unrouted means the router had NO RULE, not that the doc is fresh. Classify each by hand | | "No feature docs exist — skip Phase 2" | Mark task completed with reason. NEVER silently omit | | "Module unchanged — skip sub-skill" | Show file:line evidence. No proof = no skip | | "Already know what changed" | Still run git diff — partial knowledge causes missed updates | | "Phase 5 report not needed" | ALWAYS write summary report — it's the audit trail | | "I will update tasks later" | Invalid. Task status must change before/after each step in real time. | | "I'll run skills first then create tasks" | Invalid. Create/track tasks first, then execute step-skill calls. | | "I'll write the §8 TC myself, faster" | Invalid. Router only — delegate to $spec [mode=tests]; dual authors diverge. | | "[HARD] BR weakened but tests pass" | BLOCK — code-vs-spec contradiction; resolve or owner-accept, never wave through. |

IMPORTANT MUST ATTENTION create ALL 8 tasks via task tracking (or equivalent tracker) BEFORE any action and track each step live — in_progress before, completed after with evidence. IMPORTANT MUST ATTENTION router ONLY — delegate every §8 / Feature Spec / derived-index / derived technical view write; NEVER author them here — why: dual authorship diverges the spec from its generated views. IMPORTANT MUST ATTENTION every skip needs file:line evidence and a completed task with reason; run the fixed phase order — NEVER silently omit a phase.

Hookless Prompt Protocol Mirror (Auto-Synced)

Source: .claude/.ck.json + .claude/skills/shared/sync-inline-versions.md (:full blocks) + .claude/scripts/lib/hookless-prompt-protocol.cjs

[WORKFLOW-EXECUTION-PROTOCOL] [BLOCKING] Workflow Execution Protocol — MANDATORY IMPORTANT MUST CRITICAL. Do not skip for any reason.

Generic portability boundary: Reusable skills and protocol text stay project-neutral; project-specific conventions are discovered from docs/project-config.json and docs/project-reference/. Apply shared AI-SDD from shared/sdd-artifact-contract.md. Read docs/project-config.json and docs/project-reference/docs-index-reference.md, then open the project reference docs named there. For spec, test-case, behavior-change, public-contract, or docs/specs/ work, route through the local spec docs named by the docs index: feature-spec-reference.md, spec-system-reference.md, spec-principles.md, and workflow-spec-test-code-cycle-reference.md when specs/tests/code must stay synchronized. If either file or a required reference doc is missing or stale, auto-run $project-init (or the narrow lower-level route such as $project-config, $docs-init, $scan-all, or $scan --target=<key>) before ordinary project-specific work. Any supported AI tool may execute when this shared context and local docs are available.

DETECT: If the prompt starts with an explicit slash skill/workflow command, execute it directly. Otherwise match the prompt against the workflow catalog and skill list.
ANALYZE: Choose the best option: execute directly, invoke a skill, activate a standard workflow, or compose a custom step combination.
AUTO-SELECT: Pick the best option yourself. Do not ask the user to choose between direct execution, skill, standard workflow, or custom workflow.
ACTIVATE: For a selected workflow, call $start-workflow <workflowId>; for a selected skill, invoke that skill; for a custom workflow, sequence custom steps directly; for direct execution, proceed with the task.
CREATE TASKS: task tracking for ALL workflow/skill/custom steps before execution when the selected path has multiple steps.
PARALLELIZE: Before executing the task list, tag each task PAR (independent inputs + write set disjoint from every other PAR task) or SEQ (name the blocking dependency), group PAR tasks into waves, declare the wave plan, and spawn each wave's sub-agents in ONE message — all-return barrier per wave, fan-out one level deep unless a sub-agent's own definition authorizes further fan-out. Sequential-by-default is a defect when tasks are independent; do not parallelize shared write targets, output-consuming tasks, trivial single-file work, ordering a skill or workflow explicitly fixes, or user-approval gates.
EXECUTE: Advance per the Workflow Step Advancement & Parallel Phases rule in your context instructions — model-driven; a sub-agent completion advances a step identically to an inline call; a parallel-phase group is an all-return barrier (advance only after ALL members return, never serialize it)

Shared AI-SDD Protocol Markers

Source: .claude/skills/shared/sync-inline-versions.md

SYNC:ai-sdd-artifact-contract

AI-SDD Artifact Contract — Shared spec-driven development rules stay portable and source-owned.

Keep reusable AI-SDD principles in .claude; put repository-specific paths, commands, owners, products, and formats in project config/reference docs.

Preserve cycle: spec -> plan -> tasks -> implement -> verify -> update spec/docs.

Trace every requirement or invariant through decision, task, TC/test, source evidence, and docs/spec update.

Treat code-to-spec extraction as reference-only until accepted by the canonical spec owner.

Any supported AI tool may plan, implement, review, or verify with synced context; using multiple tools is optional.

Update .claude source first, then sync generated mirrors; do not manually edit .agents, .codex, or AGENTS.md. — why: mirrors are generated artifacts; hand-edits are overwritten on the next sync

If docs/project-config.json, root instruction files, or a required project-reference doc is missing or stale, auto-run $project-init or the narrow lower-level route before ordinary project-specific work.

Active reference: shared/sdd-artifact-contract.md in the active skills root.

SYNC:ai-sdd-artifact-contract:reminder

MANDATORY Apply shared/sdd-artifact-contract.md; keep reusable AI-SDD in .claude and local rules in project docs.
MANDATORY Code-to-spec extraction is reference-only until canonical acceptance; any supported AI tool may execute with synced context.
MANDATORY Update .claude source before syncing generated mirrors; do not manually edit .agents, .codex, or AGENTS.md.
MANDATORY Missing or stale project config, root instruction files, or required reference docs route project-specific work through $project-init or the narrow setup route automatically. [TASK-PLANNING] [MANDATORY] BEFORE executing any workflow or skill step, create/update task tracking for all planned steps, then keep it synchronized as each step starts/completes.

[LESSON-LEARNED-REMINDER] [BLOCKING] Task Planning & Continuous Improvement — MANDATORY. Do not skip.

Break work into small tasks (task tracking) before starting. Add final task: "Analyze AI mistakes & lessons learned".

Extract lessons — ROOT CAUSE ONLY, not symptom fixes:

Name the FAILURE MODE (reasoning/assumption failure), not symptom — "assumed API existed without reading source" not "used wrong enum value".
Generality test: does this failure mode apply to ≥3 contexts/codebases? If not, abstract one level up.
Write as a universal rule — strip project-specific names/paths/classes. Useful on any codebase.
Consolidate: multiple mistakes sharing one failure mode → ONE lesson.
Recurrence gate: "Would this recur in future session WITHOUT this reminder?" — No → skip $learn.
Auto-fix gate: "Could $code-review/$code-simplifier/$security-review/$lint catch this?" — Yes → improve review skill instead.
BOTH gates pass → ask user to run $learn. [CRITICAL-THINKING-MINDSET] Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80% to act. Anti-hallucination principle: Never present guess as fact — cite sources for every claim, admit uncertainty freely, self-check output for errors, cross-reference independently, stay skeptical of own confidence — certainty without evidence root of all hallucination. AI Attention principle (Primacy-Recency): Put the 3 most critical rules at both top and bottom of long prompts/protocols so instruction adherence survives long context windows. Goal-driven execution: Define success criteria first, loop until verified, and stop only when observable checks pass. Tests verify intent: Tests must protect business rules/invariants and fail when the protected intent breaks, not only mirror current behavior.

Common AI Mistake Prevention (System Lessons)

Re-read files after context compaction. Edit requires prior Read in same context; compaction wipes read state. Re-read before editing.
Grep for old terms after bulk replacements. AI over-trusts find/replace completeness. Grep full repo after bulk edits for missed refs in docs/configs/catalogs.
Check downstream references before deleting. Deletions cascade doc/code staleness. Map referencing files before removal.
After memory loss, check existing state before creating new. Compaction wipes prior-work memory. Query current state to resume — never blindly duplicate.
Verify AI-generated content against actual code. AI hallucinates APIs, class names, method signatures. Grep to confirm existence before documenting/referencing.
Trace full dependency chain after edits. Changing a definition misses downstream consumers. Trace the full chain.
When renaming, grep ALL consumer file types. Some file types silently ignore missing refs (no compile error). Search code, templates, configs, generated files.
Trace ALL code paths when verifying correctness. Code existing ≠ code executing. Trace early exits, error branches, conditional skips — not just happy path.
Update docs that embed canonical data when source changes. Docs inlining derived data (workflows, schemas, configs) go stale silently. Update all embedding docs alongside source.
Verify sub-agent results after context recovery. Background agents may finish while parent compacted — grep-verify output, don't trust assumed completion.
Cross-check full target list against sub-agent assignments. Parallel sub-agents by category miss boundary items. Reconcile union of assignments against target list before proceeding.
Sub-agents inherit knowledge only from their agent .md definition — use custom agent types, not built-in Explore. Tool adoption = permission + knowledge + enforcement (numbered workflow step).
Persist sub-agent findings incrementally, not as a final batch. Long sub-agents hit cutoffs before final write — findings lost. Instruct append-per-section to report file.
When debugging, ask "whose responsibility?" before fixing. Trace caller (wrong data) vs callee (wrong handling). Fix at responsible layer — never patch symptom site.
Test failure → record a provisional verdict before trace/edit, then investigate. Use the full five-way taxonomy: SOURCE-WRONG (production violates intent), TEST-WRONG (assertion/setup is stale), TEST-NOT-OPTIMAL (valid but fragile or low-signal test), ENVIRONMENT-BLOCKED (external state prevents a verdict), or AMBIGUOUS (intent/evidence cannot choose safely). Then trace root cause and triangulate against the governing spec (docs/specs/** if one exists) AND source. NEVER weaken an assertion, add a skip, relax a timeout, or change source merely to force green.
Grep ALL removed names after extraction/refactoring. Primary file "done" ≠ secondary files clean. Grep entire scope for every removed symbol before declaring complete.
Assume existing values are intentional — ask WHY before changing OR flagging one as a defect. Pattern-matching as "wrong" skips context. Before changing or reporting any constant/limit/flag/cutoff: read comments, git blame, the CALLER's ordering (the guarantee that makes the value correct usually lives in code running immediately BEFORE the cited line), and 2+ sibling call sites of the same convention. A doc stating WHAT without WHY is missing rationale, not proof of a missing guard — and in a validation pass, an accurate file:line citation proves the transcription, never the defect.
Verify ALL affected outputs, not just the first. One build green ≠ all green. Multi-stack changes (backend/frontend/tests/docs) require verifying EVERY output.
Evaluate fit before copying a nearby pattern. Closest example ≠ matching preconditions — verify the new context shares the same constraints, base classes, scope, lifetime.
Holistic-first debugging — resist nearest-attention trap. Don't dive into first plausible cause. List EVERY precondition (config, env vars, paths, DB, endpoints, creds, versions, DI, data). Verify each against evidence (grep/query — not reasoning). Ask "what would falsify this?" — if nothing, it's not a hypothesis. Most expensive failure: going deeper in "obvious" layer while bug sits in layer never questioned.
Surgical changes — apply the diff test (context-aware). Two modes: (1) Bug fix → every line traces to the bug; no restyling; orphan cleanup only for imports YOUR changes made unused. (2) Review/enhancement → implement improvements AND announce as "Enhancement beyond main request: [what]". Never silently scope-creep. Diff test: "Would this line exist if I wasn't asked to do X?" — if no, delete or announce.
Surface ambiguity before coding — don't pick silently. Multiple valid interpretations → present each with effort: "[Request] could mean (1) [N h], (2) [N h]. Which matters?" List scope/format/volume/constraints assumptions first. If simpler path exists, say so. Never silently pick.
[MANDATORY FIRST ACTION] ALWAYS activate a suitable skill or workflow BEFORE responding. Match task against workflow catalog + skill list; invoke via skill invocation or $start-workflow <workflowId>. NEVER answer or write code before checking. Skip = protocol violation.
Why-Review adversarial mindset — apply when reviewing any plan, decision, or design. Default SKEPTIC not VALIDATOR: steel-man a rejected alternative, invert each stated reason ("what does it sacrifice?"), stress-test top 2-3 assumptions, run pre-mortem ("ships, fails in 3 months — what breaks?"), surface 1-2 alternatives author missed. Section presence ≠ quality; quality = causal reasoning + concrete mitigations + evidence, not "it's better" or "monitor closely".
Front-load report-write in sub-agent prompts for large reviews. Many-file sub-agents hit budget before final write — findings lost. Design prompts so: (1) report-write is first explicit deliverable, (2) append per-file/section (not batched), (3) scope bounded so reads don't exhaust budget. Truncated mid-sentence with no report file → spawn narrower scope, don't retry same prompt.
After context compaction, re-verify all prior phase outcomes before continuing. Summaries describe intent, not environment state (git index, filesystem, processes). On resume, FIRST audit: git status, re-read modified files, verify filesystem. Every "completed" claim is an untested hypothesis until evidence confirms.
OOM/memory: check row count before row size. Triage: (1) Unbounded query — no DB filter for trigger? Push filter to DB; eliminates OOM. (2) Large rows? Projection reduces proportionally. Row reduction > projection in ROI.
Assert the outcome your system OWNS, never the intermediate state your INFRASTRUCTURE owns. When testing anything asynchronous (queue/broker delivery, retries, background jobs, caches, replication), assert the final business/entity state. NEVER assert the delivery bookkeeping — consume/send status, attempt counts, last-error, row existence or counts in a broker, scheduler, or outbox/inbox table. That bookkeeping lives in shared infrastructure that ANY co-running process (a peer worker, a second replica, a leftover local container) can write, usually under a deterministic shared key, so the assertion silently tests the developer's environment instead of the system: green when run alone, flaky the instant anything else shares that broker + database. Gate question for every assertion: "would this hold no matter WHICH process did the work?" — if no, assert the converged data state instead. Corollary: process-local fault injection and in-process telemetry cannot gate work any process may perform — use them as stress amplifiers (arm → bounded window → disarm → assert convergence), never as preconditions.
Keep domain concepts out of generic/shared/infrastructure layers. Reusable layer (shared library, framework, infra module) must reference NO consumer-specific domain concept — tenant/customer/product IDs, business entities, feature rules. Leak compiles + runs → passes review silently while coupling the "reusable" layer to one consumer. Keep shared type domain-free; push domain fields/logic down into the consumer via subclass/composition. — why: a layer coupled to one consumer's domain is no longer reusable.

Agent Skills: docs-update

Install this agent skill to your local

Skill Files