메뉴
[Planning] Use when you need to auto-review a plan for validity, correctness, and best practices — recursive: review, validate findings with why-review, fix validated findings, full re-review until no findings.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
[Architecture] Use when auditing the ENTIRE project architecture and production readiness in one pass — bundles architecture-review + architecture-scalability-review + production-readiness-review at project or diff scope, then synthesizes one consolidated Architecture Health Report.
[Code Quality] Use when reviewing architecture compliance for layers, messaging, service boundaries, CQRS, repos, entity events, and data/consistency/tenancy boundaries. Universal architecture laws, coupling taxonomy and the anti-pattern catalog live in `.claude/docs/architecture-knowledge.md` (project docs always outrank it).
[Code Quality] Use when you need to review artifact quality (PBI, user story, test spec, design spec) before handoff. Supports --type={pbi|story|spec-tests|design}.
[Code Quality] Use when reviewing current changes, staged or unstaged diffs, or branch-to-branch diffs.
[Code Quality] Use when evaluating review feedback, requesting targeted code-quality review, or verifying completion claims.
[DDD Quality] Use when you need to review domain entities and value objects for DDD design quality.
| name | plan-review |
| description | [Planning] Use when you need to auto-review a plan for validity, correctness, and best practices — recursive: review, validate findings with why-review, fix validated findings, full re-review until no findings. |
Codex compatibility note:
- Invoke repository skills with
$skill-namein Codex; this mirrored copy rewrites legacy Claude/skill-namereferences.- 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_agentsubagent(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 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-reference.mdbackend-patterns-reference.md, domain-entities-reference.mdfrontend-patterns-reference.md, scss-styling-guide.md, design-system/README.mddocs/specs/ pathing, or TC format: feature-spec-reference.md, spec-system-reference.md, spec-principles.mdworkflow-spec-test-code-cycle-reference.md plus the spec docs abovespec-system-reference.md and source Feature Specs under docs/specs/integration-test-reference.mde2e-test-reference.mdcode-review-rules.md plus domain docs above based on changed filesDo 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_progresswhen step starts, setcompletedwhen 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.
Goal: Block any plan from reaching implementation unless it is hallucination-free (every existing-code claim proven at file:line) and implementation-ready (every step concrete, small enough to code from immediately) — by auto-reviewing implementation plans for validity, correctness, and best practices. Recursive: when any findings exist, validate findings with $why-review --validate-findings, fix only validated findings in plan files, and rerun the full plan review until no findings remain.
Summary: AI self-review (automatic, NOT a user interview like $plan-validate) that gates a plan before implementation.
file:line proof (Anti-Hallucination Gate); every phase must clear the "Detailed & Small Enough" granularity gate (≤5 files, ≤3h, no planning verbs) — too vague → detail it, too big → break it.plan.md + goal.md + all phase-*.md, extract requirements/steps/files/risks → Step 2 evaluate the 4 checklist groups: Validity (summary, requirements, steps, files) · Correctness (Granularity Gate + Anti-Hallucination/Code-Proof Gate + spec/TC coverage + Goal-Contract mapping) · Best Practices (YAGNI/KISS/DRY/architecture) · Completeness (risks, testing, success criteria, security, graph-dependency) → run the 11 Adversarial techniques + Anti-Bias Gate + 7 Plan Dimensions → graph-trace each modified file (when graph.db exists) → Step 3 score PASS/WARN/FAIL → Step 4 output result → Step 5 recursive validate-fix-re-review loop.$why-review --validate-findings gate BEFORE editing any plan.md/phase-*.md, fix only validated findings at the smallest responsible location, then restart the FULL review with a fresh, zero-memory sub-agent — loop until a clean pass with zero findings.Workflow:
## Plan Context$why-review --validate-findings on the plan-review report first; fix only validated actionable issues in plan files, then re-review (loop back to step 2 until zero findings, unless the repeated-blocker rule or the 5-round cap applies)Core Principle — Detailed & Small Enough:
Key Rules:
file:line proof — unverified paths, class names, or behaviors = FAIL$why-review --validate-findings, fix only validated plan issues, then re-run the FULL review from the start. Repeat this self-loop — no forced minimum, capped at 5 rounds MAX — until a complete pass finds ZERO findings.Be skeptical. Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence percentages (Idea should be more than 80%).
The success metric of every coding decision is future change cost. DRY, SRP, abstraction, design patterns, naming, layering, tests — every technique exists to serve one goal: making the next change cheaper.
Evaluating code, refactor, test, abstraction, ask: does this make next change cheaper or more expensive?
Apply this lens before invoking any specific rule, pattern, or checklist below — if a downstream rule raises change cost, this principle wins.
Default stance: SKEPTIC, not validator. Your job is to find what cannot work, not confirm what looks right.
Confirmation bias trap: After reading a well-structured plan, AI naturally finds reasons to agree. This section exists to break that loop before it produces a rubber-stamp approval.
Techniques 1-6 stress whether the plan can be built (reality, effort, scope, dependencies). Techniques 7-10 stress whether the chosen design is the right one — the decision-quality lens shared with
$why-review's rationale review. Apply both groups: a buildable plan built on the wrong decision is still a failed plan.
1. Implementation Reality Check For every phase, ask: "If a developer started implementing this right now, what is the first thing that would break?" Walk through the critical path concretely. Vague phases ("implement the service layer") that can't be traced to specific files/classes fail this check.
2. Assumption Stress Test List the top 3 implicit assumptions embedded in the plan. For each: "What if this assumption is wrong?" A valid plan survives at least 2 of its 3 assumptions being false. Common hidden assumptions: "existing code is in a known state," "no external API changes," "team has this domain knowledge."
3. Effort Reality Check For each phase marked with effort estimates: "Has similar work in this codebase been done in this timeframe? What slowed it down last time?" Plans that underestimate by 2x or more are not valid plans — they are optimistic guesses.
4. Pre-Mortem Assume the plan is implemented exactly as written and the feature is in production after 1 month. Write one concrete failure scenario that is plausible given the current plan. If you can't find one, you haven't looked hard enough.
5. Scope Creep Detector Identify any task in the plan that is NOT directly required to deliver the stated feature. "While we're here, let's also refactor X" is scope creep. Flag it.
6. Dependency Blindspot List 2-3 external dependencies (other services, APIs, data sources) the plan assumes are stable. For each: "What breaks in this plan if this dependency changes or is unavailable?" If a dependency failure is not addressed anywhere in the plan, it is a risk gap.
7. Steel-Man the Rejected Alternative For each design decision where the plan chose approach X over an alternative, argue FOR the rejected alternative as strongly as you can. Would a 10-year domain senior have chosen it? If yes, the plan's dismissal needs stronger proof than "we picked X." A decision that never names what it rejected has not been made — it has been assumed.
8. Why NOT? For every "chose X because Y" in the plan, ask what X sacrifices. Every choice has a cost; a plan that lists only the upsides of its chosen approach is hiding the trade-off, not avoiding it. Demand the named downside.
9. Unseen Alternatives Identify 1-2 viable approaches the plan does NOT mention at all. An alternative absent without exclusion reasoning is weak design coverage, not a settled decision. Name them and ask why they were not considered.
10. Pros/Cons Symmetry Count the chosen approach's stated pros vs cons. Pros outnumbering cons by more than 2:1 signals confirmation bias, not a clean design — demand the missing downsides before accepting the decision.
11. Contrarian Pass Before writing any verdict, generate at least 2 sentences arguing the OPPOSITE conclusion. If you're about to write PASS — argue for NEEDS WORK. If about to write NEEDS WORK — argue for PASS. Then decide which argument is stronger based on evidence.
Complete ALL checks before writing the final verdict (MUST ATTENTION):
If any check is incomplete → you have NOT completed the adversarial review. Go back.
Why-review relationship: Techniques 7-10 + these gate checks are the rationale lens applied DURING the review pass (does the plan's design hold up?). The separate
$why-review --validate-findingsgate runs AFTER findings exist (are the findings themselves correct before we fix them?). Both stay — they validate different things and must not be collapsed.
After plan-type detection (Phase 0), evaluate each dimension below using this reasoning pattern:
For each dimension: (1) Understand its role in the plan's domain, (2) Read the plan's claims about it, (3) Derive the actual concerns from first principles — what could go wrong if this dimension is weak? (4) Apply your knowledge of the plan's tech stack to find stack-specific gaps.
Think: Does the plan's scope match the stated goal exactly — not broader, not narrower?
Think: Can I trace how data moves through every phase of this plan?
Think: Does the plan account for everything its changes affect?
Think: What does the plan say about when things go wrong?
Think: How will a developer know if this plan's implementation is correct?
[HARD] §4 rule / §5 invariant gets a universally-quantified property test spec plus a boundary counter-case — not example tests only; (2) changed core logic is gated by a MUTATION-SCORE quality bar (a surviving mutant = a missing invariant ⇒ a killing test owed), NOT a line-coverage % target; (3) a dual-feedback + re-review step exists so each behavior-changing finding enriches BOTH the spec AND the tests and the package is re-reviewed to zero new gaps. FAIL a plan that targets a line-coverage % instead of a mutation-score bar, or that omits property/invariant test specs for its [HARD]/§5 rules.Think: Does implementing this plan require knowledge the plan doesn't surface?
Think: Does the frontmatter estimation still match the finalized plan, or did scope-locking change the cost?
bottom_up_hours = Σ phase_hours from each phase file's locked tasks/TCs and compare to current frontmatter man_days_traditional / story_points.likely_days, risk_margin_pct, min-max range per SYNC:estimation-framework. Did unknowns resolve (margin should drop) or new risks surface (margin should rise)?|delta| > 20% → frontmatter MUST be updated with reestimate_delta_pct: <signed> + 1-line reestimate_reason. Missing update = FAIL.|delta| > 50% → flag SHOULD-RESCOPE in review verdict; the plan must surface the rescope decision to the user before implementation begins.Use these dimensions to generate targeted, evidence-backed questions — not generic "add more detail" suggestions.
Perform automatic self-review of implementation plan — ensure valid, correct, follows best practices; identify anything needing fixes before proceeding.
Key distinction: AI self-review (automatic), NOT user interview like $plan-validate.
$ARGUMENTS provided -> Use that path## Plan Context section -> Use active plan pathBefore applying any checklist, read plan.md and classify the plan:
| Signal in plan | Type | Additional review focus |
|---|---|---|
| "fix", "bug", "regression", "defect" in title/description | Bugfix | Behavioral Delta Matrix (MANDATORY), preservation inventory, regression tests |
| "migrate", "schema", "database", "index" | Data/Schema | Rollback path, zero-downtime strategy, data preservation, migration idempotency |
| "auth", "permission", "security", "encrypt", "token", "RBAC" | Security | Threat modeling, attack surface, trust boundary changes, sub-agent: security-auditor |
| "performance", "latency", "cache", "N+1", "throughput" | Performance | Baseline metrics, regression risk, measurement strategy, sub-agent: performance-optimizer |
| "refactor", "extract", "rename", "restructure" | Refactor | Behavior preservation, blast radius, dangling references |
| "API", "contract", "endpoint", "consumer", "event" | Contract/Integration | Backward compatibility, consumer impact, versioning strategy |
| "infra", "CI", "pipeline", "deploy" | Infrastructure/DevOps | Rollback plan, environment parity, secrets handling |
| None of the above | Feature | Standard checklist, acceptance criteria mapping, YAGNI |
If multiple signals match, list all types and apply ALL their focus areas.
Plan type drives:
Read the plan directory:
plan.md - Overview, phases list, frontmattergoal.md - Goal Contract (when present): Original Request, Purpose, Success Criteria (required vs optional), Constraintsphase-*.md - All phase filesIf {plan-dir}/goal.md is missing, resolve plans/goals/{YYMMDD-HHmm}-{slug}/goal.md; if no Goal Contract exists at all, record No active goal — plan reviewed against plan.md requirements only.
| # | Check | Presence | Quality Depth |
|---|---|---|---|
| 1 | Has executive summary — clear 1-2 sentence description | Does a summary section exist? | Is it accurate? Does it scope the work or conceal complexity? |
| 2 | Has defined requirements section — explicit requirements listed | Does a requirements section exist? | Are requirements concrete user needs or vague technical goals? |
| 3 | Has implementation steps — actionable tasks | Are implementation steps present? | Are steps specific (file names, method names) or vague actions? |
| 4 | Has files to create/modify listing — file inventory present | Is a file listing present? | Are file paths real (verified via glob/grep)? Do they follow project conventions? |
Decision tree — apply to EACH phase:
Phase too vague? (no file paths, planning verbs, unclear actions)
→ YES → DETAIL IT: add specific file paths, exact method names, concrete actions
→ NO ↓
Phase too big? (>5 files OR >3h effort OR single step is a mini-project)
→ YES → BREAK IT: split into smaller sibling phases until each meets limits
→ NO → PASS this phase
5-Point Criteria (all must pass per phase):
| # | Criterion | PASS example | FAIL example |
|---|---|---|---|
| 1 | Steps name specific files | "Modify {source-root}/auth/login" | "Implement authentication" |
| 2 | No planning verbs | "Add validateToken() method" | "Determine the best auth approach" |
| 3 | Each step ≤30 min effort | "Add error handler to endpoint" | "Build the entire auth module" |
| 4 | Phase ≤5 files AND ≤3h | 3 files, 2h | 12 files, 8h |
| 5 | No unresolved decisions | All approaches decided | "TBD: which library to use" |
Planning verbs that trigger FAIL: "research", "determine", "figure out", "decide", "evaluate", "explore", "investigate" — these belong in investigation, not implementation plans.
Action on failure (after Findings Validation Gate passes):
Do not apply these refinements until $why-review --validate-findings returns CLEAN for the current plan-review report.
{plan-dir}/sub-plans/phase-{XX}-{name}/plan.mdWorked example:
FAILS: "Phase 2: Data Layer — Set up database models, Create repositories, Implement data access patterns. Effort: 4h, Files: ~8"
PASSES after split: "Phase 2A: Data Schema (1h, 3 files) — Create {source-root}/models/user-entity, Create {source-root}/models/session-entity, Create {migration-root}/create-users-sessions" + "Phase 2B: Repository Layer (1.5h, 3 files) — Create {source-root}/repositories/user-repository, Create {source-root}/repositories/session-repository, Register in {composition-root}"
file:line proof.| Claim type | Required proof |
|---|---|
| File path | File exists (glob/read) |
| Class/method name | Symbol grep → file:line |
| Behavior ("X calls Y") | Code evidence file:line |
| Base class / interface | Inheritance verified (grep/graph) |
FAIL triggers: unread file paths, ungrepped method names, "should be"/"probably"/"typically" language about existing code, behaviors assumed from similar projects instead of THIS codebase. Greenfield-only plans (no existing code refs) → PASS.
## Test Specifications section with TC mappings. "TBD" is valid for TDD-first mode.goal.md exists: every saved required success criterion is covered by ≥1 phase, and each phase's success criteria trace to saved criteria (or are marked supporting work with reason). FAIL if a saved required criterion has no covering phase, or the plan delivers work the Goal Contract never asked for without recorded justification. Skip with No active goal evidence when no Goal Contract exists.| # | Check | Presence | Quality Depth |
|---|---|---|---|
| 1 | YAGNI — No unnecessary features or over-engineering | Is every planned component traceable to a stated requirement? | Flag anything described as "might be useful" or added for future flexibility without a current requirement. |
| 2 | KISS — Simplest viable solution chosen | Is there a stated approach for each major step? | Could any planned abstraction be simpler with the same effect? Are there unnecessary layers, indirections, or framework choices? |
| 3 | DRY — No planned duplication of logic | Are there similar patterns described more than once? | Does the plan introduce new patterns when existing ones work? Are there repeated steps that suggest duplication at implementation time? |
| 4 | Architecture — Follows project patterns from .claude/docs/ | Does the plan reference or align with .claude/docs/ patterns? | Does it follow established patterns or deviate? Any deviations need explicit justification with rationale. |
| # | Check | Presence | Quality Depth |
|---|---|---|---|
| 1 | Risk assessment present with mitigations — risks identified with responses | Is there a risk section with at least one item? | Are mitigations specific actions (who, when, triggered by what) or vague intentions ("monitor closely")? |
| 2 | Testing strategy defined — test approach outlined | Is there a testing section or test references per phase? | Does it cover unit, integration, and edge case paths, or just "write tests"? Is the approach traceable to acceptance criteria? |
| 3 | Success criteria per phase — measurable outcomes defined | Does each phase have stated success criteria? | Are criteria measurable? Would failing them trigger a rollback, or are they aspirational targets? |
| 4 | Security considerations addressed — security concerns noted | Is there a security section or inline security notes? | Are security concerns specific to this feature's attack surface, or generic boilerplate (e.g., "use HTTPS", "validate inputs")? |
| 5 | Graph dependency check — importers of modified files are checked | If .code-graph/graph.db exists: are importers_of queries run for each modified file? | Are ALL importers checked, not just direct callers? Is the graph.db prerequisite explicitly stated? Are missed dependents flagged? |
| Status | Criteria | Action |
|---|---|---|
| PASS | All Required pass, ≥50% Recommended | Proceed to implementation |
| WARN | All Required pass, <50% Recommended | Proceed with caution, note gaps |
| FAIL | Any Required check fails | STOP - must fix before proceeding |
## Plan Review Result
**Status:** PASS | WARN | FAIL
**Reviewed:** {plan-path}
**Date:** {current-date}
### Summary
{1-2 sentence summary of plan quality}
### Checks Passed ({X}/{Y})
#### Required ({X}/{Y})
- ✅ Check 1
- ✅ Check 2
- ❌ Check 3 (if failed)
#### Recommended ({X}/{Y})
- ✅ Check 1
- ⚠️ Check 2 (missing)
### Issues Found
- ❌ FAIL: {critical issue requiring fix}
- ⚠️ WARN: {minor issue, can proceed}
### Recommendations
1. {specific fix 1}
2. {specific fix 2}
### Verdict
{PROCEED | REVISE_FIRST | BLOCKED}
When graph DB is available, verify the plan covers all affected files:
python .claude/scripts/code_graph trace <file> --direction downstream --jsonProtocol:
SYNC:double-round-trip-review+SYNC:fresh-context-review+SYNC:review-protocol-injection(all inlined above in this file).
When the review results in FAIL, WARN, or any non-zero findings, plan-review MUST run the Findings Validation Gate before editing any plan file. Only findings validated by $why-review --validate-findings may be fixed. After fixing validated actionable findings, rerun the full plan-review protocol from the first review step over the current plan. Do not spawn a fresh sub-agent just to re-review known findings before fixing them. If the restarted full review uses a sub-agent, it uses the canonical Agent template from SYNC:review-protocol-injection below and re-reads ALL plan files from scratch with ZERO memory of prior fixes.
Trigger this gate whenever the plan-review output contains any finding: FAIL, WARN, recommendation requiring a plan edit, missing evidence, unresolved risk, or implementation-blocking ambiguity. Skip this gate only when the completed review pass has zero findings.
$why-review --validate-findings against that report in the main review flow before editing plan files.$why-review --validate-findings before any fix.NEVER edit plan.md or phase-*.md to fix review findings before this gate passes. This gate validates findings; the fresh full plan-review happens only after the validated fix cycle.
When constructing the Agent call prompt for Round N (N≥2):
SYNC:review-protocol-injection template verbatimagent_type: "general-purpose" (this is a plan review, not a code review)SYNC:evidence-based-reasoning, SYNC:rationalization-prevention, SYNC:graph-assisted-investigation, SYNC:understand-code-first (omit code-specific protocols like SYNC:bug-detection, SYNC:test-spec-verification which are not applicable to plan files)"Review plan files under {plan-dir}. Validate structural completeness, code-proof anti-hallucination (every file:line claim about existing source code must exist), and adversarial simulation (imagine implementing each phase right now — what fails first?).""read plan.md and all phase-*.md files under {plan-dir}"plans/reports/plan-review-round{N}-{date}.mdAfter the sub-agent returns:
## Re-Review {N} Findings in the main report — DO NOT filter or override┌──────────────────────────────────┐
│ Round 1: Main-session review │
│ (structural checklist + basic │
│ code-proof trace) │
│ Output: PASS / WARN / FAIL │
└──────────────┬───────────────────┘
│
┌──────▼──────┐
│ ZERO │
│ FINDINGS? │──YES──→ Proceed to next workflow step
└──────┬──────┘
│ NO
┌──────▼──────────────────────────────────┐
│ VALIDATE: Run $why-review │
│ --validate-findings on the report. │
│ Only validated findings may be fixed. │
└──────┬──────────────────────────────────┘
│
┌──────▼──────────────────────────────────┐
│ FIX: Modify plan files to resolve │
│ validated actionable findings │
│ (plan.md/phase-*) │
└──────┬──────────────────────────────────┘
│
┌──────▼──────────────────────────────────┐
│ Round 2+: FULL PLAN RE-REVIEW │
│ Re-run the complete plan-review │
│ protocol from the first review step. │
│ If the protocol uses agents, spawn │
│ new agents for that restarted pass. │
└──────┬──────────────────────────────────┘
│
└──→ Loop until zero findings, repeated-blocker rule, or 5-round cap
MANDATORY — NO EXCEPTIONS after completing this skill, you MUST use ask the user directly to present these options. Do NOT skip because the task seems "simple" or "obvious" — the user decides:
[BLOCKING] This is a validation gate. MUST ATTENTION use ask the user directly to present review findings and get user confirmation. Completing without asking at least one question is a violation.
[IMPORTANT] Use task tracking to break ALL work into small tasks BEFORE starting — including tasks for each file read. This prevents context loss from long files. For simple tasks, AI must ask user whether to skip.
Critical Purpose: Ensure quality — no flaws, no bugs, no missing updates, no stale content. Verify both code AND documentation.
External Memory: For complex or lengthy work (research, analysis, scan, review), write intermediate findings and final results to a report file in
plans/reports/— prevents context loss and serves as deliverable.
Evidence Gate: MANDATORY — every claim, finding, and recommendation requires
file:lineproof or traced evidence with confidence percentage (>80% to act, <80% must verify first).
OOP & DRY Enforcement: MANDATORY — flag duplicated patterns that should be extracted to a base class, generic, or helper. Classes in the same group (same suffix, same lifecycle, same purpose) must share a common base (even if empty now — enables future shared logic and child overrides). Verify project has code linting/analyzer configured for the stack.
Behavioral Delta Matrix — MANDATORY for bugfix reviews. Produce this table BEFORE PASS/FAIL verdict. Narrative descriptions don't substitute.
Input state Pre-fix behavior Post-fix behavior Delta {condition} {current behavior} {fixed behavior} Preserved ✓ / Fixed ✓ / REGRESSION ✗ Rules: ≥3 rows · ≥1 row the bug report did NOT mention · REGRESSION delta → FAIL until a preservation test covers it (
spec-tests-template.md#preservation-tests-mandatory-for-bugfix-specs)BLOCKED until: ≥3 rows · ≥1 row outside bug report · no unmitigated REGRESSION
Graph-Assisted Investigation — MANDATORY when
.code-graph/graph.dbexists.HARD-GATE: MUST ATTENTION run at least ONE graph command on key files before concluding any investigation.
Pattern: Grep finds files →
trace --direction bothreveals full system flow → Grep verifies details
Task Minimum Graph Action Investigation/Scout trace --direction bothon 2-3 entry filesFix/Debug callers_ofon buggy function +tests_forFeature/Enhancement connectionson files to be modifiedCode Review tests_foron changed functionsBlast Radius trace --direction downstreamCLI:
python .claude/scripts/code_graph {command} --json. Use--node-mode filefirst (10-30x less noise), then--node-mode functionfor detail.
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,IntegrationEventEvent consumers Consumer,EventHandler,Subscribe,@EventListener,inboxSagas/orchestration Saga,ProcessManager,Choreography,Workflow,OrchestratorSync 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
Fresh Context Re-Review — Eliminate orchestrator confirmation bias after fixes by restarting the full review with isolated sub-agents where applicable.
Why: The main agent knows what it (or
$feature-implement) just fixed and rationalizes findings accordingly. A fresh sub-agent has ZERO memory, re-reads from scratch, and catches what the main agent dismissed. Sub-agent bias is mitigated by (1) fresh context, (2) verbatim protocol injection, (3) main agent not filtering the report.When: ONLY after a validated-finding fix cycle. A review round that finds zero issues ENDS the loop — do NOT spawn a confirmation sub-agent. A review round that finds issues triggers: validate findings → fix → full review restart from the first phase.
How:
- Start a NEW full review invocation/task breakdown; when that protocol calls for agents, spawn NEW
spawn_agenttool calls — usecode-revieweragent_type for code reviews,general-purposefor plan/doc/artifact reviews- Inject ALL required review protocols VERBATIM into the prompt — see
SYNC:review-protocol-injectionfor the full list and template. Never reference protocols by file path; AI compliance drops behind file-read indirection (seeSYNC:shared-protocol-duplication-policy)- Sub-agent re-reads ALL target files from scratch via its own tool calls — never pass file contents inline in the prompt
- Sub-agent writes structured report to
plans/reports/{review-type}-round{N}-{date}.md- Main agent reads the report, integrates findings into its own report, DOES NOT override or filter
Rules:
- SKIP fresh sub-agent when the prior full review found zero issues (no fixes = nothing new to verify)
- NEVER skip the full review restart after a fix cycle — every fix invalidates the prior verdict
- NEVER reuse a sub-agent across rounds — every fresh round spawns a NEW
spawn_agentcall- Continue until a complete full review pass has zero findings; if the same blocker repeats 3 times with no progress, escalate by asking the user directly
- Track iteration count and repeated blockers in conversation context (session-scoped, no persistent files)
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=trueand recordparentTaskId; 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_progressbefore work andcompletedimmediately 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.
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_progressbefore work andcompletedimmediately after evidence; never batch transitions.- For plan/review work, create
plans/reports/{skill}-{YYMMDD}-{HHmm}-{slug}.mdbefore 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.
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.
Sequential Thinking Protocol — Structured multi-step reasoning for complex/ambiguous work. Use when planning, reviewing, debugging, or refining ideas where one-shot reasoning is unsafe.
Trigger when: complex problem decomposition · adaptive plans needing revision · analysis with course correction · unclear/emerging scope · multi-step solutions · hypothesis-driven debugging · cross-cutting trade-off evaluation.
Format (explicit mode — visible thought trail):
Thought N/M: [aspect]— one aspect per thought, state assumptions/uncertaintyThought N/M [REVISION of Thought K]: ...— when prior reasoning invalidated; state Original / Why revised / ImpactThought N/M [BRANCH A from Thought K]: ...— explore alternative; converge with decision rationaleThought N/M [HYPOTHESIS]: ...then[VERIFICATION]: ...— test before actingThought N/N [FINAL]— only when verified, all critical aspects addressed, confidence >80%Mandatory closers: Confidence % stated · Assumptions listed · Open questions surfaced · Next action concrete.
Stop conditions: confidence <80% on any critical decision → escalate by asking the user directly · ≥3 revisions on same thought → re-frame the problem · branch count >3 → split into sub-task.
Implicit mode: apply methodology internally without visible markers when adding markers would clutter the response (routine work where reasoning aids accuracy).
Deep-dive: see
$sequential-thinkingskill (.claude/skills/sequential-thinking/SKILL.md) for worked examples (API design, debugging, architecture), advanced techniques (spiral refinement, hypothesis testing, convergence), and meta-strategies (uncertainty handling, revision cascades).
Project Reference Docs Gate — Run after task-tracking bootstrap and before target/source file reads, grep, edits, or analysis. Project docs override generic framework assumptions.
- Identify scope: file types, domain area, and operation.
- Read
docs/project-config.jsonfirst — the project's machine-readable map. It is the single source of truth for THIS repo (modules/paths, framework + search keywords, test/E2E/integration run-commands, design system, architecture rules, workflow patterns); ground exact paths, run-commands, and conventions on it before investigating, planning, or coding — never assume framework defaults (CLAUDE.md+ reference docs are derived from it). If it — or the docs index,lessons.md,CLAUDE.md,AGENTS.md, or any required reference doc — is missing or stale, auto-run$project-initor the narrow route ($project-config,$docs-init,$scan-all,$scan --target=<key>,$claude-md-init) first; if Codex mirrors orAGENTS.mdare stale, ask the user to run$sync-codex(never auto-run it).- Required docs by trigger: always
docs/project-reference/lessons.md; doc lookupdocs-index-reference.md; reviewcode-review-rules.md; backend/CQRS/APIbackend-patterns-reference.md; domain/entitydomain-entities-reference.md; frontend/UIfrontend-patterns-reference.md; styles/designscss-styling-guide.md+design-system/design-system-canonical.md; integration testsintegration-test-reference.md; E2Ee2e-test-reference.md; feature docs/specsfeature-spec-reference.md+spec-system-reference.md+spec-principles.md; behavior/public-contract/spec-test-code syncworkflow-spec-test-code-cycle-reference.md; derived spec index/ERD/reimplementation guidesspec-system-reference.md+ source Feature Specs underdocs/specs/; architecture/new areaproject-structure-reference.md.- Read every required doc, then before target work state:
Reference docs read: ... | Not applicable: ....Ready when: scope evaluated,
docs/project-config.jsonconsulted, required docs checked/read or setup route completed,lessons.mdconfirmed, citation emitted.
Understand Code First — HARD-GATE: Do NOT write, plan, or fix until you READ existing code.
- Search 3+ similar patterns (
grep/glob) — citefile:lineevidence- Read existing files in target area — understand structure, base classes, conventions
- Run
python .claude/scripts/code_graph trace <file> --direction both --jsonwhen.code-graph/graph.dbexists- Map dependencies via
connectionsorcallers_of— know what depends on your target- Write investigation to
.ai/workspace/analysis/for non-trivial tasks (3+ files)- Re-read analysis file before implementing — never work from memory alone. — why: long context drifts from the file; the file is ground truth
- NEVER invent new patterns when existing ones work — match exactly or document deviation. — why: divergent patterns fragment the codebase and slow every future reader
BLOCKED until:
- [ ]Read target files- [ ]Grep 3+ patterns- [ ]Graph trace (if graph.db exists)- [ ]Assumptions verified with evidence
Validated-Finding Fix + Full Re-Review Loop — Re-review is triggered by a validated finding fix cycle, not by a round number. Review purpose:
review → validate findings → fix validated findings → full re-reviewuntil a complete review pass finds no issues. A clean review ENDS the loop — no further rounds required.aka Self-Review Convergence Loop. The name is historical — there is NO 2-round cap; "double-round-trip" only means a validated-finding fix cycle forces at least one fresh re-review. It runs until a clean pass, bounded by the 5-round ceiling below.
Round cap — 5 rounds MAX (a ceiling, NEVER a target). A clean pass ENDS the loop immediately at ANY round — round 1 included; the cap never obliges you to keep spinning. Hitting round 5 with validated findings still open → STOP and escalate by asking the user directly with the still-open findings listed; NEVER emit a silent "good enough" PASS on cap exhaustion, and NEVER let the cap substitute for the clean-review requirement. The 3-repeated-no-progress blocker rule stays an EARLIER exit — escalate at whichever trips first.
Universal scope (any new output/judgment): any newly produced output or judgment gets ≥1 self-review; any new judgment gets ≥1
$why-review --validate-findingspass; anything flagged to re-check is re-checked ≥1 time — before that output is treated as final. This loop is the default convergence contract for ANY work-producing skill, not review skills only.Routing invariant (author-facing): a skill that validates findings MUST route them through
$why-review --validate-findings(the terminal validator) — NEVER fork an inline finding-validation. Routing through why-review is what makes the finding-survival bar and this loop apply; theverify-review-validate-coveragesensor enforces this exact route mechanically.Round 1: Main-session review. Read target files, build understanding, note issues. Output findings + verdict (PASS / FAIL).
Decision after Round 1:
- No issues found (PASS, zero findings) → review ENDS. Do NOT spawn a fresh sub-agent for confirmation.
- Issues found (FAIL, or any non-zero findings) → run the active review skill's findings-validation gate first; for review skills the default gate is
$why-review --validate-findings <report-path>. Fix only validated findings, then restart the full review protocol from the beginning with a fresh task breakdown.Fresh full re-review after every fix cycle: Re-run the whole review protocol over the current full target. When sub-agents are part of that protocol, spawn NEW
spawn_agentcalls — never reuse prior agents. Reviewers re-read ALL files from scratch with ZERO memory of prior rounds. SeeSYNC:fresh-context-reviewfor the spawn mechanism andSYNC:review-protocol-injectionfor the canonical Agent prompt template. Each fresh full review must catch:
- Cross-cutting concerns missed in the prior round
- Interaction bugs between changed files
- Convention drift (new code vs existing patterns)
- Missing pieces that should exist but don't
- Subtle edge cases the prior round rationalized away
- Regressions introduced by the fixes themselves
Loop termination: After each full re-review, repeat the same decision: clean → END; issues → validate findings → fix → restart from the first review phase. Continue until a complete review pass finds zero issues, capped at 5 rounds. Escalate by asking the user directly at whichever comes first: the same validated finding repeats for 3 full invocations with no progress · a fix requires product/owner input · round 5 completes with validated findings still open. NEVER loop past 5 rounds, and NEVER convert cap exhaustion into a PASS.
Rules:
- A clean Round 1 ENDS the review — no mandatory Round 2
- NEVER fix unvalidated findings; validate first using the caller's validation gate
- Every surviving finding must additionally clear the finding-survival bar defined in why-review's Findings Validation Routine (a deliberately higher bar than the generic act-gate — "keep this finding?" is a stricter question than "act on this evidence?"); a finding below the bar is demoted or dropped, not kept
- NEVER skip the full re-review after a fix cycle (every fix invalidates the prior verdict)
- NEVER reuse a sub-agent across rounds — every iteration that uses sub-agents spawns NEW Agent calls
- Main agent READS sub-agent reports but MUST NOT filter, reinterpret, or override findings
- The 5-round cap NEVER replaces the clean-review requirement — it bounds runaway looping, it does not authorize shipping an un-clean review; a clean pass ends the loop early at any round, and cap exhaustion escalates rather than passes
- Enforce the round cap of 5 alongside the 3 repeated-no-progress blocker rule; both are escalation triggers, neither is a completion criterion
- Track recursive invocation count and repeated blockers in conversation context (session-scoped)
- Final verdict must incorporate ALL rounds executed
Report must include
## Round N Findings (Fresh Sub-Agent)for every round N≥2 that was executed.
Review Protocol Injection — Every fresh sub-agent review prompt MUST embed 11 protocol blocks VERBATIM. The template below has ALL 11 bodies already expanded inline. Copy the template wholesale into the Agent call's
promptfield at runtime, replacing only the{placeholders}in Task / Round / Reference Docs / Target Files / Output sections with context-specific values. Do NOT touch the embedded protocol sections.Why inline expansion: Placeholder markers would force file-read indirection at runtime. AI compliance drops significantly behind indirection (see
SYNC:shared-protocol-duplication-policy). Therefore the template carries all 11 protocol bodies pre-embedded.
code-reviewer — for code reviews (reviewing source files, git diffs, implementation)general-purpose — for plan / doc / artifact reviews (reviewing markdown plans, docs, specs)spawn_agent({
description: "Fresh Round {N} review",
agent_type: "code-reviewer",
prompt: `
## Task
{review-specific task — e.g., "Review all uncommitted changes for code quality" | "Review plan files under {plan-dir}" | "Review integration tests in {path}"}
## Round
Round {N}. You have ZERO memory of prior rounds. Re-read all target files from scratch via your own tool calls. Do NOT trust anything from the main agent beyond this prompt.
## Protocols (follow VERBATIM — these are non-negotiable)
### Spec ↔ Tests ↔ Code Triangulation
DO THIS FIRST — before any per-protocol check below. The review target is the WHOLE PACKAGE, not the diff alone: load the behavior's spec (§3 ACs / §4 BRs / §8 TCs), its tests, and the changed code TOGETHER, and reason about their mutual consistency BEFORE judging any one in isolation.
1. Locate all three faces: the Feature Spec section(s) governing the changed behavior, the tests that guard it, and the production code that implements it. A missing face is itself a finding (SPEC-GAP / TEST-GAP / DEAD-SPEC).
2. Triangulate pairwise — every disagreement is a finding; classify which face is wrong:
- code vs spec: behavior the code does that no §3/§4/§8 rule describes → CODE-EXTRA or SPEC-STALE; a [HARD] §4 rule or §5 invariant with no enforcing code path → CODE-WRONG.
- tests vs spec: a §8 TC with no test, or a test asserting behavior no TC/rule names → TEST-GAP or SPEC-SILENT.
- tests vs code: a changed code path with no covering test → TEST-GAP; a test that still passes against a deliberately broken invariant → WEAK-TEST (apply the mutation thinking in Bug Detection).
3. Hidden-rule capture: any invariant the code enforces but the spec never states (SPEC-SILENT) MUST be surfaced as a finding to add into §3/§4/§8 AND guarded with a test — the enrichment loop, never a silent pass.
4. Only after the three faces agree — or every disagreement is logged as a finding — proceed to the per-protocol checks below; when enrichment adds spec/test content, re-review the package against the enriched spec.
NEVER mark review PASS while any spec/test/code face disagrees without a logged finding. The diff is the entry point; the package is the unit of judgment.
### Evidence-Based Reasoning
Speculation is FORBIDDEN. Every claim needs proof.
1. Cite file:line, grep results, or framework docs for EVERY claim
2. Declare confidence: >80% act freely, 60-80% verify first, <60% DO NOT recommend
3. Cross-service validation required for architectural changes
4. "I don't have enough evidence" is valid and expected output
BLOCKED until: Evidence file path (file:line) provided; Grep search performed; 3+ similar patterns found; Confidence level stated.
Forbidden without proof: "obviously", "I think", "should be", "probably", "this is because".
If incomplete → output: "Insufficient evidence. Verified: [...]. Not verified: [...]."
### Bug Detection
MUST check categories 1-4 for EVERY review. Never skip.
1. Null Safety: Can params/returns be null? Are they guarded? Optional chaining gaps? .find() returns checked?
2. Boundary Conditions: Off-by-one (< vs <=)? Empty collections handled? Zero/negative values? Max limits?
3. Error Handling: Try-catch scope correct? Silent swallowed exceptions? Error types specific? Cleanup in finally?
4. Resource Management: Connections/streams closed? Subscriptions unsubscribed on destroy? Timers cleared? Memory bounded?
5. Concurrency (if async): Missing await? Race conditions on shared state? Stale closures? Retry storms?
6. Stack-Specific: Check the configured language/runtime pitfalls and framework-specific failure modes discovered from local code.
Classify: CRITICAL (crash/corrupt) → FAIL | HIGH (incorrect behavior) → FAIL | MEDIUM (edge case) → WARN | LOW (defensive) → INFO.
### Design Patterns Quality
Priority checks for every code change:
1. DRY via OOP: Same-suffix classes (*Entity, *Dto, *Service) MUST share base class. 3+ similar patterns → extract to shared abstraction.
2. Right Responsibility: Logic in LOWEST layer (Entity > Domain Service > Application Service > Controller). Never business logic in controllers.
3. SOLID: Single responsibility (one reason to change). Open-closed (extend, don't modify). Liskov (subtypes substitutable). Interface segregation (small interfaces). Dependency inversion (depend on abstractions).
4. After extraction/move/rename: Grep ENTIRE scope for dangling references. Zero tolerance.
5. YAGNI gate: NEVER recommend patterns unless 3+ occurrences exist. Don't extract for hypothetical future use.
Anti-patterns to flag: God Object, Copy-Paste inheritance, Circular Dependency, Leaky Abstraction.
### Logic & Intention Review
Verify WHAT code does matches WHY it was changed.
1. Change Intention Check: Every changed file MUST serve the stated purpose. Flag unrelated changes as scope creep.
2. Happy Path Trace: Walk through one complete success scenario through changed code.
3. Error Path Trace: Walk through one failure/edge case scenario through changed code.
4. Acceptance Mapping: If plan context available, map every acceptance criterion to a code change.
5. Tests Verify Intent: For test/spec changes, verify tests name the protected business rule or invariant and would fail if that intent breaks.
6. Migration Test Exclusion: Do not write tests for migration code. Schema/data migrations are one-time execution paths, not core application logic.
NEVER mark review PASS without completing both traces (happy + error path).
### Test Spec Verification
Map changed code to test specifications.
1. Identify the project's test/spec format from existing docs, test-case files, BDD feature files, or spec folders.
2. Every changed code path MUST map to a corresponding test case/spec (or flag as "needs test case").
3. New functions/endpoints/handlers → flag for test spec creation.
4. Migration files are excluded from test/spec creation; schema/data migrations are one-time execution paths, not core application logic.
5. If spec evidence fields exist, verify they point to actual code (file:line, not stale references).
6. Verify each meaningful test case names the business intent/invariant; flag behavior-only cases that only mirror implementation details.
7. Auth/data changes → verify corresponding authorization and data-state test cases exist.
8. If no specs exist for a changed path → log the gap and recommend the project's test-spec workflow.
NEVER skip test mapping. Untested code paths are the #1 source of production bugs.
### Behavioral Delta Matrix
MANDATORY for any bugfix review. Produce input-state × pre-fix × post-fix × delta table BEFORE writing verdict.
- Minimum 3 rows; include at least one row OUTSIDE the original bug report.
- Any "REGRESSION" delta → review returns FAIL until a preservation test is added.
- Narrative descriptions do NOT substitute for the matrix.
Example rows (external-record sync fix):
| Input | Pre-fix | Post-fix | Delta |
| --------------------- | ------- | ------------------------- | ---------- |
| Record exists (valid) | Reused | Always recreated → orphan | REGRESSION |
| Record missing (404) | Error | Recreated | Fixed |
### Fix-Layer Accountability
NEVER fix at the crash site. Trace the full flow, fix at the owning layer. The crash site is a SYMPTOM, not the cause.
MANDATORY before ANY fix:
1. Trace full data flow — Map the complete path from data origin to crash site across ALL layers (storage → backend → API → frontend → UI). Identify where bad state ENTERS, not where it CRASHES.
2. Identify the invariant owner — Which layer's contract guarantees this value is valid? Fix at the LOWEST layer that owns the invariant, not the highest layer that consumes it.
3. One fix, maximum protection — If fix requires touching 3+ files with defensive checks, you are at the wrong layer — go lower.
4. Verify no bypass paths — Confirm all data flows through the fix point. Check for direct construction skipping factories, clone/spread without re-validation, raw data not wrapped in domain models, mutations outside the model layer.
BLOCKED until: Full data flow traced (origin → crash); Invariant owner identified with file:line evidence; All access sites audited (grep count); Fix layer justified (lowest layer that protects most consumers).
Anti-patterns (REJECT): "Fix it where it crashes" (crash site ≠ cause site, trace upstream); "Add defensive checks at every consumer" (scattered defense = wrong layer); "Both fix is safer" (pick ONE authoritative layer).
### Rationalization Prevention
AI skips steps via these evasions. Recognize and reject:
- "Too simple for a plan" → Simple + wrong assumptions = wasted time. Plan anyway.
- "I'll test after" → RED before GREEN. Write/verify test first.
- "Already searched" → Show grep evidence with file:line. No proof = no search.
- "Just do it" → Still need task tracking. Skip depth, never skip tracking.
- "Just a small fix" → Small fix in wrong location cascades. Verify file:line first.
- "Code is self-explanatory" → Future readers need evidence trail. Document anyway.
- "Combine steps to save time" → Combined steps dilute focus. Each step has distinct purpose.
### Graph-Assisted Investigation
MANDATORY when .code-graph/graph.db exists.
HARD-GATE: MUST run at least ONE graph command on key files before concluding any investigation.
Pattern: Grep finds files → trace --direction both reveals full system flow → Grep verifies details.
- Investigation/Scout: trace --direction both on 2-3 entry files
- Fix/Debug: callers_of on buggy function + tests_for
- Feature/Enhancement: connections on files to be modified
- Code Review: tests_for on changed functions
- Blast Radius: trace --direction downstream
CLI: python .claude/scripts/code_graph {command} --json. Use --node-mode file first (10-30x less noise), then --node-mode function for detail.
### Understand Code First
HARD-GATE: Do NOT write, plan, or fix until you READ existing code.
1. Search 3+ similar patterns (grep/glob) — cite file:line evidence.
2. Read existing files in target area — understand structure, base classes, conventions.
3. Run python .claude/scripts/code_graph trace <file> --direction both --json when .code-graph/graph.db exists.
4. Map dependencies via connections or callers_of — know what depends on your target.
5. Write investigation to .ai/workspace/analysis/ for non-trivial tasks (3+ files).
6. Re-read analysis file before implementing — never work from memory alone.
7. NEVER invent new patterns when existing ones work — match exactly or document deviation.
BLOCKED until: Read target files; Grep 3+ patterns; Graph trace (if graph.db exists); Assumptions verified with evidence.
## Reference Docs (READ before reviewing)
- `.claude/docs/development-rules.md` — canonical development rules, code-quality guidelines, and pre-commit checklist
- docs/project-reference/code-review-rules.md
- {skill-specific reference docs — e.g., integration-test-reference.md for integration-test-review; backend-patterns-reference.md for backend reviews; frontend-patterns-reference.md for frontend reviews}
## Target Files
{explicit file list OR "run git diff to see uncommitted changes" OR "read all files under {plan-dir}"}
## Output
Write a structured report to plans/reports/{review-type}-round{N}-{date}.md with sections:
- Status: PASS | FAIL
- Issue Count: {number}
- Critical Issues (with file:line evidence)
- High Priority Issues (with file:line evidence)
- Medium / Low Issues
- Cross-cutting findings
Return the report path and status to the main agent.
Every finding MUST have file:line evidence. Speculation is forbidden.
`
})
{placeholders} in Task / Round / Reference Docs / Target Files / Output sections with context-specific contentcode-reviewer agent_type for code reviews and general-purpose for plan / doc / artifact reviewsAI 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. Before changing a constant, limit, flag, wording, or pattern, read nearby context and history. Surface ambiguity before acting — don't pick silently. Multiple valid interpretations require an explicit question or stated assumption with risk. Keep shared guidance role-relevant. Universal guidance must help every receiving skill or agent; code-specific obligations belong only in code-specific protocols.
Severity Rubric — Classify every finding by consequence, not by how easy it is to fix. One scale across all reviews so a "High" means the same thing everywhere.
Severity Action Definition CRITICAL Block merge Silent runtime failure, data corruption, validation bypass, security hole HIGH Must fix Incorrect behavior, invariant gap, architectural violation MEDIUM Should fix Design debt, maintainability, likely future bug LOW Nice to fix Convention, documentation, minor clarity Score-based skills map their numeric scale onto these tiers — do not invent a parallel vocabulary:
- 0-2 criterion scoring (e.g. production-readiness-review):
0= CRITICAL/HIGH (criterion unmet, blocks production readiness),1= MEDIUM (partial, should fix),2= pass (no finding).- Two-axis scoring (e.g. performance-review, impact × likelihood): map the resulting cell to the nearest tier — high-impact + high-likelihood → CRITICAL/HIGH; low-impact OR low-likelihood → MEDIUM/LOW.
A finding's tier drives the gate: CRITICAL/HIGH must be resolved or explicitly accepted by the owner before PASS; MEDIUM/LOW may ship with a tracked follow-up.
Goal Contract Satisfaction Loop — Persist the user goal in an external file, execute against it, and loop review/fix until every saved required criterion passes or a blocker escalates. Bounded closed loop — NEVER open-ended autonomous exploration.
- Resolve the active goal (in order): active plan
goal.md→plans/goals/{YYMMDD-HHmm}-{slug}/goal.md→ create a new Goal Contract from the current user request (template:.claude/templates/goal-contract-template.md).