[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}.
설치
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
이 SKILL.md는 매우 커서 SkillsMP가 여기에는 첫 섹션만 미리 보여줍니다.GitHub에서 보기
name
artifact-review
description
[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}.
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
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: Review an artifact (PBI, design spec, story, test spec) for completeness and quality so reviewed artifacts are complete, evidence-backed, and ready for handoff without missing assumptions or acceptance gaps.
Summary:
Purpose: review ONE artifact (PBI · user story set · test spec · design spec) for completeness + quality so it ships evidence-backed and handoff-ready — no missing assumptions, no acceptance gaps. Default stance = SKEPTIC, not presence-checker: sections that exist but hold weak/untestable content are worse than missing ones — they breed false confidence.
Main steps (in order): (1) Identify type — dispatch on --type={pbi|story|spec-tests|design}, infer if omitted; (2) Adversarial Mindset — run ALL 6 techniques (steel-man rejected alternatives · stress-test 3 assumptions · AC-testability · pre-mortem · unseen alternatives · contrarian pass) + clear the Anti-Bias Gate before any verdict; (3) Type checklist — score Required + Recommended; (4) M1-M7 gate (BLOCKING, ALL types); (5) Readability checklist; (6) Output per-type template (verdict + Required/Recommended tallies + coverage/AC matrix); (7) Validated-fix + full re-review loop — validate findings → fix → restart full review until clean.
Type dispatch + verdict: each --type has its own Required/Recommended checklist and output template — verdict = PASS (all Required + ≥50% Recommended) | WARN (all Required, <50% Recommended) | FAIL (any Required fails).
M1-M7 gate (ALL types, BLOCKING): any M1-M5 or M7 violation forces NEEDS WORK citing the mandate ID + exact section/line; exempt source identifiers inside evidence carriers ([Source:], **Evidence**, CoveredBy, legacy IntegrationTest, frontmatter, mermaid) — flag tech leakage only in narrative/AC/scenario prose. — why: carriers are the correct home for class/path/test names; flagging them is a false finding. M7 (business-visibility) is judged on each case's BODY via the demo test, NOT its prose — a tech-free-sounding case about a consumer/sync/handler passes M1 and STILL fails M7. — why: M1 governs vocabulary, M7 governs subject matter; the gap between them is how business specs rot.
Validated-fix loop: before fixing, invoke $why-review --validate-findings <report-path> on the review report FIRST (validate-before-fix discipline, at parity with $plan-review) — NEVER edit the artifact to resolve findings before this gate returns CLEAN. Then fix only validated findings, do not confirm-in-place, restart the FULL review (fresh general-purpose sub-agent — artifacts are NOT code), loop until a clean pass — clean review ENDS the loop.
Workflow:
Identify — What artifact type is being reviewed
Checklist — Apply type-specific quality criteria
Verdict — READY or NEEDS WORK with specific items
Key Rules:
Use type-specific checklists
Every NEEDS WORK item must be actionable
Focus on completeness — never block on stylistic preferences
Be skeptical. Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence percentages (Idea should be more than 80%).
First Principle — Easy to Change
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.
When evaluating code, refactor, test, or abstraction, ask:
does this make next change cheaper or more expensive?
Name real enemies in findings: coupling, hidden state, duplicated
knowledge, unclear intent, irreversible decisions exposed too early.
Simpler design easy to change beats sophisticated design that isn't.
Apply this lens before invoking any specific rule, pattern, or checklist
below — if downstream rule would raise change cost, this principle wins.
Adversarial Review Mindset (NON-NEGOTIABLE)
Default stance: SKEPTIC challenging artifact quality and completeness, not confirming presence of sections.
Presence-quality confusion trap: Artifact with all required sections LOOKS complete. But sections that exist yet contain weak, ambiguous, or untestable content are worse than missing sections — they create false confidence. This section forces quality challenge beyond existence checks.
Adversarial Techniques (apply ALL before concluding)
1. Steel-Man the Alternatives
Before accepting chosen approach in any design artifact: argue FOR strongest rejected alternative as vigorously as possible. Would a senior domain expert seriously consider it? If yes — artifact's dismissal needs stronger justification.
2. Assumption Stress Test
List the 3 biggest assumptions embedded in artifact. For each: "What if this is wrong?" An artifact that breaks when 2 of its 3 core assumptions fail is fragile. Flag unaddressed failure modes.
3. Acceptance Criteria Testability
For each acceptance criterion: "Can a QA engineer write a specific automated test for this — without asking clarifying questions?" If not — AC is ambiguous. Flag it. Vague ACs ("feature works correctly") are NOT acceptance criteria.
4. Pre-Mortem
Assume artifact is implemented exactly as written and feature fails in production within 3 months. Write the most plausible failure scenario. If you can't find one, look harder — every implementation has a failure mode.
5. Unseen Alternatives
Identify 1-2 approaches NOT mentioned in artifact. Genuinely not considered, or considered and excluded without documented reasoning? Missing alternatives without exclusion reasoning = incomplete analysis.
6. Contrarian Pass
Before writing any verdict, generate at least 2 sentences arguing the OPPOSITE conclusion. Then decide which argument is stronger based on evidence.
Forbidden Patterns
"Required sections present" → Presence ≠ quality. What's IN them?
"Acceptance criteria are defined" → Are they TESTABLE? Name the automated test for each.
"Scope is well-defined" → What is explicitly OUT of scope? If nothing is out of scope, the scope is undefined.
"Alternatives were considered" → Were they real alternatives, or strawmen set up to lose?
"Looks complete" → What specific failure mode is NOT addressed?
Anti-Bias Gate (MANDATORY before finalizing verdict)
Steel-manned at least one rejected alternative
Identified 3 hidden assumptions and stress-tested them
Verified each AC is unambiguously testable (can write automated test without clarification)
Ran pre-mortem (one concrete production failure scenario)
Identified at least 1 unexamined alternative (not in artifact)
Generated at least 2 sentences arguing the opposite verdict
If any box is unchecked → adversarial review incomplete. Go back.
Type-Specific Checklists (--type dispatch)
Select the checklist + output template by artifact type. Pass --type={pbi|story|spec-tests|design} to force a type; if omitted, infer from the artifact (Phase 1 "Identify"). Each type scores Required (all must pass) + Recommended (≥50% should pass) → verdict PASS (all Required + ≥50% Recommended) | WARN (all Required, <50% Recommended) | FAIL (any Required fails).
--type
Artifact
Output template
pbi
Product Backlog Item
PBI Review Result
story
User story set
Story Review Result (+ AC Coverage Matrix)
spec-tests
Test specification
Test Spec Review Result (+ Coverage Matrix)
design
Design spec
Artifact Review
PBI Review (--type=pbi)
#
Check
Presence
Quality Depth
1
Problem statement is clear — the problem being solved is described in concrete terms
Is a problem statement present? Is it 2+ sentences?
Is the problem scoped correctly? Could it be framed differently to lead to a different (simpler) solution? Are symptoms confused with root cause?
2
Acceptance criteria are testable and measurable — each AC can be verified by a test
Are ACs present? Do they use measurable language?
Can a QA engineer write an automated test for EACH AC without clarification? Are they specific enough to catch regressions? Vague ACs ("feature works correctly") are not acceptance criteria.
3
Scope is well-defined (what's in and out) — both in-scope and out-of-scope items are explicitly listed
Is an in/out scope list present? Does it have both sides?
Are out-of-scope items specific enough to prevent scope creep? Is anything ambiguously in/out? A scope that says nothing is out of scope is an undefined scope.
4
Dependencies are identified — all external dependencies the PBI relies on are listed
Is a dependencies section present? Does it list items?
Are ALL dependencies listed (technical, data, service, team)? Are "can-parallel" items truly safe to parallelize, or do they share a shared resource?
5
Business value is articulated — the why behind the PBI is stated in terms of user or business outcome
Is business value described?
Is the value quantified or just stated? Does it connect to a user outcome, not just a feature delivery? "Users can now do X" is better than "we implemented feature Y".
6
Priority is assigned — the PBI has an explicit priority level
Is a priority level assigned?
Is priority justified with data (RICE/MoSCoW), or arbitrary? Is it consistent with other PBIs in the same sprint? A PBI that is "high priority" without justification is unranked.
User Story Review (--type=story)
#
Check
Presence
Quality Depth
1
Follows GIVEN/WHEN/THEN format — the story uses the structured BDD format
Are all three parts (GIVEN, WHEN, THEN) present?
Are all 3 parts present AND meaningful? Or is GIVEN trivial ("Given a user exists")? A GIVEN that describes no precondition adds no value.
2
Is independent (not dependent on other stories) — the story can be implemented without requiring another story first
Is independence stated or inferable?
Would descoping other stories prevent this story from being implemented? Implicit dependencies are as blocking as explicit ones.
3
Is estimable (team can size it) — the team has enough information to assign story points
Is the story sized or estimable based on content?
Does the team have enough info to estimate? Is "can't estimate" a sign of missing AC? If it can't be sized, it's not ready for sprint.
4
Is small enough for one sprint — the story fits within a single sprint's capacity
Is the story sized at ≤8 story points or scoped to one sprint?
Could this be split further? Stories >8SP should always be split. A story that "could fit" in a sprint but requires multiple sub-systems is likely too large.
5
Has acceptance criteria — the story defines measurable conditions for completion
Are acceptance criteria present?
Are criteria testable? Would they catch a bug if the feature works in 9/10 cases? ACs that only describe the happy path are incomplete.
Design Spec Review (--type=design)
#
Check
Presence
Quality Depth
1
All component states covered (default, hover, active, disabled, error, loading) — spec defines visual behavior for all interaction states
Are all 6 states defined?
Are edge-case states (error, loading) as fully designed as the default state, or sketched? An undesigned error state will be improvised in implementation.
2
Design tokens specified (colors, spacing, typography) — specific token values are called out, not ad-hoc values
Are token references present instead of raw values?
Are tokens from the project's token system, or are new values introduced? New values outside the token system break design consistency silently.
3
Responsive behavior defined — how the component adapts across breakpoints is documented
Are breakpoint behaviors defined?
Are ALL breakpoints covered, or only desktop and mobile? Tablet-specific layouts are the most frequently omitted. Are content truncation / overflow behaviors specified?
Are requirements specific (WCAG level, contrast ratio) or vague ("should be accessible")? Vague accessibility notes produce non-compliant implementations. Is keyboard navigation flow defined?
5
Interaction patterns documented — animations, transitions, and user interaction flows are specified
Are interaction behaviors described?
Are timing and easing values specified? Is behavior defined for both forward and reverse interactions (e.g., open AND close)? Unspecified interactions are implemented inconsistently.
6
Linked from the Feature Spec — the design-spec path is recorded in the governing Feature Spec frontmatter design_spec: key so the spec stays the navigable hub
Is the design-spec path present in the Feature Spec frontmatter design_spec: (or mockup:) key?
Does the recorded path resolve to THIS design-spec, and does the design-spec deepen the spec's tech-agnostic §6 interaction surface (same UX-role view names / observable states) rather than diverge from it? An unlinked design-spec is an orphan — FAIL. Contract: SYNC:ui-intent-layer (inlined in this skill).
Test Spec Review (--type=spec-tests)
[BLOCKING] Readdocs/project-reference/spec-principles.md — use Section 5 (Test Case Registry — TC ID format, priorities, minimum coverage) and Section 6 (Evidence Format) as review criteria in addition to the checklist below.
[BLOCKING] Tech-agnostic check: flag framework/product/language/design-pattern names in a TC's behavioral prose as findings (per spec-principles.md §3). Source paths, class names, and test identifiers (e.g. {File}::{Method}) are CORRECT in evidence fields (**Evidence**, CoveredBy, legacy IntegrationTest, [Source:]), frontmatter, and Mermaid — never flag those.
[BLOCKING] Business-oriented TCs / one-to-many cardinality: Each TC must read as a business / user-story acceptance scenario, not a per-class/per-method technical unit. Flag any TC that appears split or narrowed just to mirror code structure (e.g. one TC per handler/component when the user-observable behavior is the same) — that breaks M1/M5 and the business orientation. One business TC is expected to be covered by many annotation-tagged tests across components/services; that one-to-many shape is correct and is NEVER a finding. Canonical contract: .claude/skills/shared/tc-format.md → TC ↔ Test Code Cardinality.
Required (all must pass)
#
Check
Presence
Quality Depth
1
TC ID format — All TCs follow TC-{FEATURE}-{NNN} format
Do all TCs use the TC-{FEATURE}-{NNN} pattern?
Are IDs unique per TC? Does the FEATURE code match the actual feature?
2
Story coverage — Every user story has at least one corresponding TC
Does every story ID appear in at least one TC?
Does each TC actually test the story behavior, or does it just reference the story ID in a comment?
3
AC coverage — Every acceptance criterion has a test case
Is every AC traceable to at least one TC?
Does each AC have a TC that would FAIL if the AC is violated? Single-test-per-AC misses edge cases.
4
Happy path — Each story has at least one happy path TC
Is a happy path TC present per story?
Does the happy path TC verify the full end-to-end scenario, or just a happy-path stub?
5
Error path — Each story has at least one error/failure TC
Is an error/failure TC present per story?
Does the error TC verify the exact error response (code + message), not just that an error occurred?
6
No duplicate TCs — No two TC IDs describe the same business scenario
Are all TC IDs unique with distinct business scenarios?
Flag near-duplicate TCs (same scenario, trivially different input). NOTE: this is about duplicate TCs — multiple test methods sharing one TC is the expected one-to-many shape, NEVER a duplicate.
7
Testable assertions — Each TC has clear expected result (not vague "should work")
Does each TC have a specific expected result?
Is each assertion specific enough to catch regressions? Would it pass if the return value is wrong?
8
Business intent / invariant guarded — Each TC names the rule it protects
Does every meaningful TC include Business Intent / Invariant Guarded?
Would the TC fail if that business rule/invariant breaks, or does it only mirror implementation details?
9
Authorization TCs — At least 1 TC per story verifying unauthorized access is rejected
Is an authorization TC present per story?
Does the authorization TC test a realistic access scenario, not just "wrong role → 403 without body check"?
10
TC format completeness — Every TC has Related Behaviors anchor table and CoveredBy: field
Does every TC include a Related Behaviors anchor table and CoveredBy: field? Legacy IntegrationTest: is migration input only.
For Tested-status TCs, is CoveredBy: populated with ≥1 covering test link {TestFile}::{MethodName}, a test-filter expression, or an approved manual-QC marker — not Untested? A TC may list several covering tests; never require exactly one.
11
Preservation Tests (bugfix context) — When fixing a bug, at least 1 TC verifies the pre-fix behavior is no longer reproducible
If this is a bugfix: is there a TC that would have CAUGHT the bug before the fix?
Does the preservation TC assert the exact broken behavior (not just "no exception")?
12
Invariant / Property TCs (Spec-Loop rule 1) — an Invariant/Property TC category exists, and EACH [HARD] §4 rule / §5 invariant has ≥1 universally-quantified property TC plus a boundary counter-case
Is there a property/invariant TC category, and does every [HARD]/§5 rule appear in it?
Are properties universally quantified ("for ALL inputs in range X, Y holds") with a boundary counter-case that would FAIL if the invariant breaks — not a single happy example?
13
§6 Interaction surface present (UI-bearing specs) — the governing Feature Spec carries a non-empty §6 interaction surface (6.2 View Inventory + 6.3 Navigation Map + 6.4 Key UI States + 6.5 Per-Story Interaction Flow); a backend-only (no-UI) feature instead states the skip reason in §6
For a UI-bearing feature, are §6.2–6.5 present and non-empty (views by UX role, navigation, observable states, per-story flow)? For a backend-only feature, is the skip reason stated?
Is each UI behavior cross-referenced to US-/OP-/BR-, and is the prose M1-clean — naming ZERO frameworks/routes/URLs/CSS/component classes (those belong only in the linked design-spec)? An empty §6 on a UI feature, or framework/route/CSS leakage in §6 prose, is a FAIL. Contract: SYNC:ui-intent-layer (inlined in this skill).
[BLOCKING] Spec-Loop property coverage (--type=spec-tests): The reviewer MUST verify the Invariant/Property TC category exists and that every [HARD] §4 rule / §5 invariant maps to ≥1 universally-quantified property TC plus a boundary counter-case (example-only coverage is a finding). A missing property category — or any [HARD]/§5 rule with only example TCs and no property TC — is a blocking artifact finding (Required check #12 FAIL → NEEDS WORK).
Recommended (≥50% should pass)
#
Check
Presence
Quality Depth
1
Edge cases — Boundary values, empty inputs, max limits tested
Are edge case TCs listed?
Are these the RIGHT edge cases? Do they cover the 3 most likely production failure modes for this feature?
Are security TCs present for auth, authz, and input validation?
Do security TCs attempt realistic attack vectors (SQLi, over-posting, privilege escalation) not just "invalid token → 401"?
5
Seed data TCs — If feature needs reference data, TCs verify data exists and seeder runs correctly
If reference data is needed, does a seed data TC exist (or N/A)?
If present, does the TC assert the exact seeded data shape, not just that the seeder ran without error?
6
Data migration TCs — If schema changes exist, TCs verify data transforms correctly, rollback works, no data loss
If schema changes exist, does a migration TC exist (or N/A)?
If present, does the TC verify rollback behavior and zero data loss, not just forward migration success?
7
Test data requirements specified — the data setup needed to run each test is documented
Are test data requirements stated per test?
Is test data specific enough to create fixtures without guessing? Vague data requirements ("a valid user") will cause test setup divergence across environments.
8
GIVEN/WHEN/THEN format used — tests follow the structured BDD format
Are all tests written in GIVEN/WHEN/THEN?
Are the THEN clauses assertions on observable outcomes, or on internal state? Tests asserting on internal state are brittle and break on refactoring.
M1-M7 Compliance Gate (BLOCKING — applies to ALL artifact types)
Contract: See .claude/skills/shared/sdd-artifact-contract.md → "AI-SDD Mandates (M1-M7)". This review enforces M6: any artifact (PBI, story, design spec, test spec) that violates M1-M5 or M7 MUST receive a NEEDS WORK verdict that names the violated mandate ID and cites the exact section + line. Passing an M1-M5/M7 violation makes this review itself defective. (M6 binds THIS review, not the artifact — which is why the artifact-facing set reads M1-M5 and M7, never "M1-M6".)
Carriers are EXEMPT from M1/M2 — source identifiers are CORRECT inside [Source: ...], **Evidence**, **CoveredBy**, legacy **IntegrationTest** fields, YAML frontmatter, and ```mermaid ``` blocks. Only flag leakage in narrative prose (descriptions, AC/scenario text, rule statements). Banned prose token list: docs/project-reference/spec-principles.md §3.2.
M1 — Tech-agnostic prose. FAIL if narrative prose, headings, summaries, or AC/scenario text name a framework/product, a language-native type, or a product/design-pattern class name (banned-token list in spec-principles.md §3.2). Cite the section + leaked token.
M2 — No source code in prose. FAIL if prose expresses behavior as a class/method/file-path/namespace used as a noun (e.g. "call the create-async method") instead of the business operation (e.g. "create the record"). Source identifiers belong only in evidence carriers. Cite the section + line.
M3 — Abstract-IDs-first traceability. FAIL if a requirement/rule/AC/test-case lacks a logical ID (FR-/BR-/OP-/TC-), OR has a logical ID but no [Source: namespace/service/id] abstract-anchor evidence, OR the [Source:] evidence uses physical code coordinates or repository-root paths instead of a stack-portable abstract anchor, OR the anchor is treated as the primary citation. [Source: namespace/service/id] abstract-anchor evidence is REQUIRED and KEPT — SECONDARY to the logical ID, never the spine and never removed; physical coordinates live only in the provenance sidecar.
M4 — Unambiguous, observable criteria. FAIL if AC/expected-result prose uses vague language ("handle appropriately", "process normally", "as needed"), OR two engineers could implement it differently while both claiming conformance, OR no observable completion state / named error condition exists. (Reinforces the AC-testability technique above.)
M5 — Rebuild-from-artifact. FAIL if a competent team with ZERO codebase knowledge could not re-implement the described behavior on a different stack from the artifact alone (it relies on reading source to be understood). Cite the section + the missing detail.
M7 — Business-visibility (business-tree artifacts only). Apply the demo test to each case's BODY: "what would a stakeholder SEE change?" — no answer → FAIL as TECHNICAL-ONLY. Every Given must be a state a user could arrange, every When an action a user could take, every Then an outcome a user could see. FAIL a case whose When is an invocation (a handler runs, a consumer receives, a job fires, a model is inspected, data syncs) or whose Then asserts a schema/type/nullability/column/call-count rather than a business outcome. Judge the BODY, never the title or ID — a business-sounding title routinely fronts an invocation-shaped When. Cite the case ID + the offending When/Then clause.
🔴 M1 vs M7 — the distinction this gate exists for. M1 governs vocabulary; M7 governs subject matter. A technical case written in impeccably tech-free prose satisfies M1 while violating M7 — and that gap is the single most common way business specs rot: each bugfix adds one more tech-free-sounding case about a consumer, a sync, or a load path, every M1 check passes, and the business tree fills with cases no one can demo. ⚠️ Passing a case because its prose is clean is the exact failure this box catches. ⚠️ Conversely, do NOT fail a case merely for containing a technical-sounding noun: if a user or QC can demo the outcome, it is business — M7 asks what the case is ABOUT, not which words it uses.
If ANY box fails → verdict is NEEDS WORK; list each violated mandate ID with its concrete section/line citation in the Action Items.
Readability Checklist (MUST ATTENTION evaluate)
Before approving, verify code is easy to read, easy to maintain, easy to understand:
Schema visibility — If function computes a data structure (object, map, config), a comment should show output shape so readers don't trace the code
Non-obvious data flows — If data transforms through multiple steps (A → B → C), a brief comment should explain the pipeline
Self-documenting signatures — Function params should explain their role; flag unused params
Magic values — Unexplained numbers/strings should be named constants or have inline rationale
Naming clarity — Variables/functions should reveal intent without reading the implementation
Output Format (per --type)
Pick the template matching --type. All templates lead with a Status/Verdict and ### Required/### Recommended tallies; the pbi/story/spec-tests shapes add their own evidence sections (preserved verbatim from the former refine-review, story-review, tdd-spec-review skills).
Validated Fix + Full Re-Review (MANDATORY when fixes are applied)
Protocol:SYNC:double-round-trip-review + SYNC:fresh-context-review + SYNC:review-protocol-injection (all inlined above in this file).
Do not spawn a fresh sub-agent just to re-review the same finding set before fixing it. If the artifact needs work, fix actionable findings first, then restart the full artifact review over the current artifact. When that restarted review uses a fresh general-purpose sub-agent, use the canonical Agent template from SYNC:review-protocol-injection above. Artifact reviews (PBI, story, design spec, test spec) are NOT code — use agent_type: "general-purpose", not "code-reviewer". When constructing the Agent call prompt:
Copy the Agent call shape from the SYNC:review-protocol-injection template verbatim
Set agent_type: "general-purpose"
Embed the full verbatim body of these SYNC blocks (inlined above in this skill file): SYNC:evidence-based-reasoning, SYNC:rationalization-prevention, SYNC:understand-code-first (omit code-specific protocols like SYNC:bug-detection, SYNC:design-patterns-quality, SYNC:fix-layer-accountability which are not applicable to artifact files)
Set the Task as "Run a full fresh artifact review over the current {artifact-type} after fixes were applied. Focus on: implicit assumptions, missing coverage of edge cases / error scenarios, unverified cross-references, completeness gaps only visible on second reading, whether acceptance criteria are truly testable and measurable, and regressions introduced by fixes."
Set Target Files as the explicit artifact file path(s)
Set report path as plans/reports/artifact-review-rerun{N}-{date}.md
After sub-agent returns:
Read the sub-agent's report
Integrate findings as ## Re-Review {N} Findings in the main report — DO NOT filter or override
If NEEDS WORK: fix actionable artifact findings, then restart the full artifact review from the beginning
Repeated blocker cap: if the same blocker repeats across 3 full invocations with no progress, escalate by asking the user directly
Final verdict must incorporate findings from ALL review passes that actually ran
IMPORTANT Task Planning Notes (MUST ATTENTION FOLLOW)
Always plan and break work into many small todo tasks using task tracking
Always add a final review todo task to verify work quality and identify fixes/enhancements
Bulk Multi-Artifact Sweeps
For bulk multi-artifact review (10+ artifacts at once), use $changes-review — its Systematic Review Protocol categorizes the set and fires parallel sub-agents.
AI Agent Integrity Gate (NON-NEGOTIABLE)
Completion ≠ Correctness. Before reporting ANY work done, prove it:
Grep every removed name. Extraction/rename/delete touched N files? Grep confirms 0 dangling refs across ALL file types.
Ask WHY before changing. Existing values are intentional until proven otherwise. No "fix" without traced rationale.
Verify ALL outputs. One build passing ≠ all builds passing. Check every affected stack.
Evaluate pattern fit. Copying nearby code? Verify preconditions match — same scope, lifetime, base class, constraints.
New artifact = wired artifact. Created something? Prove it's registered, imported, and reachable by all consumers.
[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 ATTENTION ask user whether to skip.
Prerequisites:MUST ATTENTION READ before executing:
OOP & DRY Enforcement: MANDATORY IMPORTANT MUST ATTENTION — flag duplicated patterns that should be extracted to a base class, generic, or helper. Classes in the same group or suffix (ex *Entity, *Dto, *Service, etc...) MUST ATTENTION inherit a common base (even if empty now — enables future shared logic and child overrides). Verify project has code linting/analyzer configured for the stack.
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).
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.
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.
Evidence-Based Reasoning — Speculation is FORBIDDEN. Every claim needs proof.
Cite file:line, grep results, or framework docs for EVERY claim
Declare confidence: >80% act freely, 60-80% verify first, <60% DO NOT recommend
Cross-service validation required for architectural changes
"I don't have enough evidence" is valid and expected output
BLOCKED until:- [ ] Evidence file path (file:line) - [ ] 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: [...]."
Understand Code First — HARD-GATE: Do NOT write, plan, or fix until you READ existing code.
Search 3+ similar patterns (grep/glob) — cite file:line evidence
Read existing files in target area — understand structure, base classes, conventions
Run python .claude/scripts/code_graph trace <file> --direction both --json when .code-graph/graph.db exists
Map dependencies via connections or callers_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
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-review until 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-findings pass; 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; the verify-review-validate-coverage sensor enforces this exact route mechanically.
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_agent calls — never reuse prior agents. Reviewers re-read ALL files from scratch with ZERO memory of prior rounds. See SYNC:fresh-context-review for the spawn mechanism and SYNC:review-protocol-injection for 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.
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_agent tool calls — use code-reviewer agent_type for code reviews, general-purpose for plan/doc/artifact reviews
Inject ALL required review protocols VERBATIM into the prompt — see SYNC:review-protocol-injection for the full list and template. Never reference protocols by file path; AI compliance drops behind file-read indirection (see SYNC: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_agent call
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)
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 prompt field 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.
general-purpose — for plan / doc / artifact reviews (reviewing markdown plans, docs, specs)
Canonical Agent Call Template (Copy Verbatim)
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.
`
})
Rules
DO copy the template wholesale — including all 11 embedded protocol sections
DO replace only the {placeholders} in Task / Round / Reference Docs / Target Files / Output sections with context-specific content
DO choose code-reviewer agent_type for code reviews and general-purpose for plan / doc / artifact reviews
DO NOT paraphrase, summarize, or skip any protocol section
DO NOT pass file contents inline — the sub-agent reads via its own tool calls so it has a fresh context
DO NOT reference protocols by file path or tag name — the bodies are already embedded above
DO NOT introduce placeholder markers for the protocols — they must stay literally expanded
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. 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.
[BLOCKING] Capture a tech-agnostic UI/UX intent layer in every UI-bearing spec — a reader must be able to visualize how the feature works without naming any technology. When the feature has a user interface, the spec MUST ATTENTION carry an interaction-surface section so the application — not just its API — can be rebuilt on any stack:
View Inventory — list each view/screen by its UX ROLE and purpose (e.g. "list of items", "item editor", "confirmation step") and what information it presents. Describe by role, never by an implementation name.
Navigation Map — how a user moves between views: entry points, transitions, and exits. Trace how this surface connects to neighboring features already in the system.
Key observable UI States — the distinct states a user can observe per view (empty, loading, populated, error, success, permission-denied, etc.) — described as what the user perceives, not how it is rendered.
Per-story interaction flow — for each user story, the step-by-step click/action path from intent to outcome, cross-referenced to the logical IDs the spec already owns (US-/OP-/BR-).
Couple to the companion design artifact — keep deep visual fidelity (layout, tokens, pixel detail) OUT of the spec; it lives in the linked design-spec/mockup. Record that companion's path in the spec frontmatter so the spec stays the navigable hub.
M1-clean (NON-NEGOTIABLE): the prose names ZERO frameworks, routes/URLs, CSS, or component-class names — only roles, information, states, and flows. Technology detail belongs in the companion design artifact, never here.
Skip ONLY when the feature is backend-only (no UI) — state that reason explicitly in the section.
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
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).
Required sections: Original Request, Purpose, Success Criteria (checkboxes; mark required vs optional), Constraints, Evidence Required, Iteration Log, Goal Satisfaction matrix.
Before work: read the active goal and map planned work to saved success criteria — execution serves the saved criteria, never chat memory alone.
After execution/verification: append an Iteration Log entry — result, evidence references (file:line, command output, report path), remaining gaps.
Review gate: emit a Goal Satisfaction matrix — | Success Criterion | Evidence | Status | with PASS/FAIL/BLOCKED. Overall PASS requires every required criterion PASS.
Loop rule (retry): required criterion FAIL → validate the gap is real → fix → re-review only the affected criteria. Stop cleanly when all required criteria PASS.
Escalation rule (stop): two consecutive iterations with no criterion progressing, or a blocker needing user input → mark the criterion BLOCKED with a user-facing reason and escalate. NEVER loop indefinitely.
Skip rule: tiny conversational tasks may skip the goal file ONLY with a recorded one-line reason. User-accepted gate skips are recorded in the goal file with reason and scope.
Security: NEVER store secrets, tokens, credentials, or private customer data in goal files — store evidence references and redact sensitive values.
Blocked until: active goal resolved (or skip reason recorded) · saved success criteria read before edits · iteration evidence appended after execution · Goal Satisfaction matrix emitted before any PASS verdict.
Trade-Off Interrogation Gate — ALWAYS ask these THREE questions before ANY verdict, score, finding, or recommendation — about the thing under review AND about every recommendation YOU make. — why: naming a benefit without its price is an endorsement, not a review; the costliest trade-offs are the ones nobody wrote down.
Is there any trade-off? Name what it SACRIFICES. "None" / "pure win" is an unfinished analysis, NOT an answer — to claim none, state which dimensions you checked and why each is unaffected: future change cost · complexity · performance/latency · memory/cost · coupling · reversibility · migration burden · operational load · blast radius · security posture · testability · team skill/ramp · delivery time · UX.
Is it worth it? Weigh gain against sacrifice EXPLICITLY — what is gained (with a metric) · what it costs · WHO pays · WHEN it comes due — then emit WORTH IT / NOT WORTH IT / UNCLEAR. "Better" with no metric and no cost FAILS this question. NOT WORTH IT → withdraw or replace the recommendation, never keep it as-is.
Is the trade-off material enough to CONFIRM WITH THE USER? A material trade-off is the user's call, never yours. MATERIAL when ANY holds: irreversible / one-way door (data migration, public contract, storage format, vendor lock-in) · cost shifted onto someone else (another team, ops/on-call, future maintainer, end user) · one quality attribute traded for another (correctness↔speed, security↔convenience, latency↔cost, simplicity↔flexibility) · a boundary crossed (client↔server tier, service contract, event contract, shared library) · a high-consequence path (auth, money, data integrity, breaking change, High/Medium residual risk) · the worth-it verdict is UNCLEAR.
MATERIAL → STOP and confirm by asking the user directly BEFORE the verdict stands — state the trade-off, both options, what each sacrifices, and your recommendation. NOT material → record it inline with a one-line justification and proceed.