[Code Quality] Use when you need to review integration tests for assertion quality, bug protection, repeatability, and test-spec traceability โ AND verify the review target (changed production code) has test coverage (integration-first) with specโtestโcode alignment.
Standardmรครig ist der Prompt ausgewรคhlt, der zuerst die Quelle prรผft. Sie kรถnnen zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prรผfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich fรผr eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fรผgen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prรผfen und installieren.
Ein direkter Befehl รผberspringt den Prรผf-Prompt. Prรผfen Sie die Quelle, bevor Sie ihn ausfรผhren.
[Code Quality] Use when you need to review integration tests for assertion quality, bug protection, repeatability, and test-spec traceability โ AND verify the review target (changed production code) has test coverage (integration-first) with specโtestโcode alignment.
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 catalog)
docs/project-reference/*
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.
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: Ensure the review target (changed production code) is covered by tests that protect real business behavior with correct data assertions, infinite repeatability, and spec alignment โ verifying every behavior change has test coverage (integration-first, unit fallback) so that specs โ tests โ code stay aligned (spec-driven development).
Summary:
The review target is the CHANGE, not the test files: collect BOTH changed production code AND changed test files โ Gates 1-6 judge test quality, Gate 7 maps every behavior-changing production file to a covering test (integration-first; unit only with recorded justification) and a spec TC. An uncovered changed behavior is a HIGH finding minimum.
Read the handler/service source (and feature docs) BEFORE judging any assertion. FAIL smoke-only, existence-only (not-null), dead (always-true), copy-paste, and DI-resolution-only tests; require unique IDs + async polling + 3 consecutive green runs to call tests repeatable/verified.
Gate 6 three-way sync uses the source-of-truth hierarchy (feature docs > test-spec docs > impl code > test code): NEVER fix a test to match broken code, NEVER self-resolve a three-way conflict โ escalate via a direct user question.
Don't just report gaps โ Phase 5 WRITES the missing test (and runs $spec [mode=tests] for SPEC-GAPs), then a full fresh re-review runs after every validated fix cycle until a clean pass returns 0 CRITICAL/0 HIGH.
Scope: The FULL change set โ changed production code AND changed test files โ from uncommitted changes (default), user-specified files, or a user-specified diff (branch/PR). The review target is never "just the test files".
MUST collect BOTH changed production code AND changed test files โ coverage of the change is part of the review, not an optional extra
MUST verify every behavior-changing production change maps to a covering test โ integration test FIRST; unit test ONLY with recorded justification (Gate 7)
MUST treat an uncovered changed behavior as a HIGH finding minimum โ fix by writing the missing test in Phase 5, not just reporting
MUST verify specโtestโcode alignment for changed code, not only for existing tests โ a changed behavior with no TC in spec docs is a spec gap finding
MUST read handler/service source BEFORE judging any test assertions
MUST flag smoke-only tests (no-exception-only checks) as FAIL
MUST flag DI-resolution-only tests (resolve + not-null) as FAIL โ NOT integration tests
MUST verify tests use unique IDs per run (infinitely repeatable)
MUST use async polling/retry for ALL DB assertions โ async delays are norm
MUST flag repository-created or repository-mutated test data that bypasses real use cases and can leave invalid state
MUST require 3 consecutive successful suite/project runs before declaring integration tests verified/idempotent
NEVER accept assertions that always pass regardless of handler correctness
NO smoke/fake/useless tests โ every test MUST execute actual operations and verify data state
docs/project-reference/integration-test-reference.md โ Integration test patterns, fixture setup, seeder conventions, lessons learned (MUST READ before reviewing)
First Principle โ Easy to Change
The success metric of every coding decision is future change cost.
DRY, SRP, abstraction, design patterns, naming, layering, tests โ 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 โ if downstream rule would raise change cost, this principle wins.
Phase 0: Scope Detection
Classify BEFORE any gate review. Route wrong โ waste all effort.
Signal
Classification
Action
No user-specified files
Uncommitted changes
Run git diff --name-only (staged + unstaged) to collect scope โ BOTH production code AND test files
User specifies files/diff (branch, PR, etc.)
Explicit scope
Use provided list/diff directly โ still split into production vs test files
10+ test files
Large scope
Parallel sub-agents grouped by module
1-9 test files
Normal scope
Single review pass
0 test files BUT production code changed
Coverage-gap review
Gate 7 IS the review โ map every changed behavior to existing tests; uncovered behavior = finding. Do NOT exit
0 changes at all
Empty target
Ask user for explicit scope via a direct user question
The review target is the CHANGE, not the test files. Changed test files are reviewed for quality (Gates 1-6); changed production files are checked for coverage and spec alignment (Gate 7). Both halves are mandatory.
Search for test reference docs โ NEVER hardcode paths. Grep for integration-test-reference, test-patterns, integration-test-guide near changed test files to discover project-specific conventions before starting gate review.
The 7 Quality Gates
Gates 1-6 apply per changed/target TEST file. Gate 7 applies to the CHANGE SET โ every behavior-changing production file must map to a covering test and a spec TC.
Gate 1: Assertion Value โ "Would this catch the bug?" (MUTATION-SCORE gate)
Think: If a single line of the handler's core logic were changed (a > flipped to >=, a field assignment removed, a boolean negated), would at least one assertion FAIL? If NONE โ FAIL. This is the mutation-testing question, made automatic.
#1 AI failure: hallucination assertions โ look real, verify nothing.
Operationalized โ run the project's mutation tool (PRIMARY). "Would this catch the bug?" is exactly the mutation-score question. Mechanize it instead of eyeballing it:
Discover the configured mutation tool from docs/project-config.json, dependency manifests, and CI config. Common per stack: Stryker (JS/TS, .NET โ StrykerNet), PITest (Java/JVM), mutmut or cosmic-ray (Python). Cite the local config or command if one exists.
Run it scoped to the CHANGED handler/service (mutate only the production files in the review target โ never the whole repo) against the covering tests from Gate 7.
Read the surviving-mutant report. Each surviving mutant = a missing invariant = an assertion gap โ a HIGH finding minimum (CRITICAL when the mutated line touches authorization, money, or data integrity).
Fix in Phase 5 by WRITING the killing test โ the assertion or property that fails on that mutant. Re-run until the changed code's mutants are killed (or each survivor has a recorded justification, e.g. equivalent mutant). Raising the line-coverage number is NOT a fix โ coverage that executes a line without asserting its effect leaves the mutant alive.
Manual fallback (when NO mutation tool is configured or addable). Apply the single-mutation thought experiment by hand: read the handler source, then for each core-logic line ask "if I deleted or inverted this, which assertion fails?" If the answer is NONE for any business-critical line โ FAIL. Prefer recommending the stack-appropriate mutation tool as a harness add so the gate becomes automatic next time.
PASS: Every changed core-logic line is killed by โฅ1 assertion โ no surviving mutant on the changed code (mutation tool), or the manual single-mutation check finds a failing assertion for each (fallback) โ AND the Mutation Probe Ledger (below) is recorded in the report with a KILLED/SURVIVOR verdict per changed core-logic line. No ledger โ no PASS, regardless of how clean the eyeball check felt.
FAIL:
A surviving mutant on a changed core-logic line with no killing test (or no recorded equivalent-mutant justification)
No-exception as ONLY assertion
Not-null without content check
Assertions on fields handler doesn't modify
Dead assertions: x >= 0 where x always >= 0, count >= 0, string not-empty on required fields
Verify: Run the mutation tool on the changed handler โ list surviving mutants โ each survivor is a missing assertion to write. When no tool is available: read handler source โ list fields/branches it changes โ check at least one assertion would fail if each were mutated.
Recorded artifact โ Mutation Probe Ledger (REQUIRED, non-skippable, BOTH paths). "I checked it mentally" is not evidence. Gate 1 cannot be marked PASS without this ledger written into the review report โ it is the proof the probe ran, identical in obligation whether a tool ran or the manual fallback did:
SURVIVOR โ finding (or recorded equivalent-mutant justification)
Rules: (1) every changed core-logic line gets a row โ no sampling, no "representative subset". (2) Tool path: rows come from the surviving-mutant report; manual fallback: rows come from the line-by-line thought experiment โ same table, same columns. (3) A SURVIVOR row with no killing assertion is a HIGH finding minimum (CRITICAL on auth/money/data-integrity lines) UNLESS it carries a written equivalent-mutant justification. (4) An empty or absent ledger = Gate 1 FAIL (not "skipped") โ the gate is unproven, so it cannot pass.
Gate 2: Data State โ "Does it check the database?"
Think: Does this test prove the database changed, or just that no exception occurred?
PASS: After command, test queries DB and asserts specific entity field values.
FAIL:
Only checks return value, never verifies DB state
Checks existence (not-null) without field values
Missing async polling on side-effect assertions
Exception: Smoke-only ONLY when side effect truly unobservable. MUST include explicit justification comment.
ALWAYS use async polling/retry for data assertions. Event handlers, bus consumers, background jobs run async โ data may not be immediately available.
Gate 3: Repeatability โ "Can I run this 100 times?"
Think: If this test runs N times in a shared database, does it get noisier each run? Would run #2 fail?
FAIL: Hardcoded IDs, hardcoded business keys without unique suffix, teardown/cleanup, ordering dependency, seeders without existence check, or direct repository setup that creates state users could not create through real use cases.
Verify: Repeatability is only proven when the relevant suite/project passes 3 consecutive runs without resetting data. One green run is not enough.
Gate 4: Domain Logic โ "Does test match handler?"
Think: Did I read the handler source? Do I know which exact fields it writes? Do assertions check those fields โ and ONLY those fields?
PASS: Assertions match what handler ACTUALLY does (verified by reading source). Covers primary business rule. Validation paths tested.
FAIL: Assertions on untouched fields (copy-paste), missing primary side-effect assertion, event handler tests that never trigger the event.
Verify: Grep handler class โ read it โ list what it does โ compare with assertions.
Also check:
Authorization: test verifies both authorized AND unauthorized access paths?
Coverage: happy path + validation failure + DB state check (3 tests minimum)
Gate 5: Spec Traceability โ "Is this tracked?"
Think: Can I trace TC-XXX-NNN from test annotation โ spec docs โ feature docs in one unbroken chain?
PASS: Test has a TestSpec annotation linking to a TC ID that exists in spec docs. The test method name need NOT match the TC, and many test methods may legitimately carry the same TC (one business TC โ many tests across components/services โ the join key is the test-spec annotation, not the method name; see tc-format.md โ TC โ Test Code Cardinality).
FAIL (WARN, not BLOCK): Missing annotation, orphaned TC ID (annotation points to a TC absent from spec docs), or spec says "Planned" but test exists. NOT a finding: several tests sharing one TC, or a method name that differs from the TC โ those are the expected one-to-many shape.
Think: Have I read ALL 3 sources? Where exactly do they disagree? Does evidence support a verdict, or must I escalate?
Hardest gate. Identify discrepancy, classify using source-of-truth hierarchy โ NEVER silently pick winner. Always state the resolved source with file:line evidence โ why: a winner picked without evidence hides bugs.
Source of Truth Hierarchy (highest โ lowest)
Priority
Source
Why
1 (Highest)
Feature docs (docs/specs/โฆ/Section 8 TCs)
Business intent โ defines WHAT must happen
2
Test-spec docs (docs/specs/)
TC scenarios derived from feature docs โ defines HOW to verify
3
Implementation code (handler/entity/service)
What WAS built โ may reflect intentional evolution not yet in docs
4 (Lowest)
Integration test code
What IS being tested โ most likely to be wrong or stale
Rule: Docs win over code. Code wins over tests. Feature docs win over test-spec docs.
Conflict Classification
Pattern
Feature Doc
Impl Code
Test Code
Verdict
Action
All agree
โ
โ
โ
PASS
None
Stale docs
โ
โ
โ
Docs lag code
Flag docs for $docs-update; test is correct
Wrong test
โ
โ
โ
Test wrong
Fix test assertions to match code + docs
Code bug
โ
โ
โ
Code has bug
Report as BUG โ do NOT fix test to match code
Test + code diverge from docs
โ
โ
โ
Code bug + wrong test
Fix test to match docs; report code bug
Three-way conflict
โ
โ
โ
ESCALATE
Cannot self-resolve โ a direct user question
CRITICAL rules:
NEVER fix a test to match broken code โ that hides bugs
NEVER assume docs are wrong without evidence they were intentionally superseded
NEVER self-resolve a three-way conflict โ always escalate via a direct user question
"Stale docs" verdict requires BOTH code AND test to agree โ one source never enough
When escalating, include: TC ID, what each source says, evidence found
Test code: Read test method โ arrange, execute, assert
Compare each pair with file:line evidence for each source.
PASS: All three agree. WARN: Minor wording, same semantic. FAIL: Semantic disagreement on field/rule/outcome. ESCALATE: All three differ and evidence cannot resolve.
Gate 7: Change Coverage โ "Is every changed behavior tested AND specced?"
Think: For each behavior-changing production file in the review target, which test would FAIL if this change were broken? If NONE โ coverage gap. Which spec TC describes this behavior? If NONE โ spec gap.
This gate makes the skill verify the REVIEW TARGET has coverage โ not merely review tests that happen to exist.
Protocol:
Collect changed production files from the review target (Phase 0 scope): commands, queries, handlers, entities, services, event handlers, consumers, controllers, frontend services/stores with business logic.
Filter to behavior-changing files. Exclude: migrations (one-time execution paths), generated code, pure renames/formatting, config-only, DI registration-only changes. Record each exclusion with reason.
Find covering tests per changed behavior โ use graph (query tests_for <fn>, trace <file> --direction both) plus grep for the handler/class name under test directories. A test COVERS a change only if it exercises the changed path and asserts the changed outcome โ read the test; name match alone is NOT coverage.
Apply test-type priority: integration test FIRST (subcutaneous CQRS through real DI, data-state assertions). Unit test is an acceptable fallback ONLY when integration coverage is infeasible (pure function/calculation logic, no observable data state, no DI path) โ record the justification per fallback.
Check spec alignment for the change โ existence AND correctness. Each changed behavior must map to a TC in spec docs (feature doc Section 8 / test-spec docs). Finding a TC is NOT enough: READ the mapped TC and confirm it describes the CURRENT behavior. New behavior with no TC, or a TC that exists but still describes the OLD/superseded behavior โ spec gap (spec-driven development violation). A behavior is only fully covered when a covering test exercises it AND a non-stale ยง8 TC documents it โ so this correctness re-check applies to COVERED rows too, never just to GAP rows.
Coverage Mapping Table (MANDATORY output):
Rows are keyed by changed production behavior, not by TC. A behavior is COVERED when โฅ1 covering test exercises it โ list ALL covering tests in the column when several apply. One Spec TC may legitimately appear across multiple rows and be covered by many tests (one TC โ many tests, 1:N). Do NOT expect or require one test per TC, and do NOT flag a TC reused across rows as a duplicate (see tc-format.md โ TC โ Test Code Cardinality).
Changed File / Behavior
Spec TC
Covering Test(s)
Test Type
Verdict
{file:line โ behavior}
TC-X-NNN / MISSING
{test file:method}[, โฆone or more] / NONE
integration / unit (justified) / โ
COVERED / COVERED-UNIT / GAP / SPEC-GAP
Verdicts:
COVERED โ integration test exercises the changed path with data-state assertions AND the mapped ยง8 TC describes the CURRENT behavior. A covering test whose mapped TC is stale is NOT COVERED โ record it as SPEC-GAP.
COVERED-UNIT โ unit test covers it, integration infeasible, justification recorded, and the mapped ยง8 TC is current (same stale-TC rule applies).
GAP (FAIL) โ no test would fail if the change broke. Severity: HIGH minimum; CRITICAL when the change touches authorization, money, or data integrity. Fix in Phase 5 by WRITING the missing test (integration-first) โ reporting alone does not clear this gate
SPEC-GAP (FAIL) โ behavior has no TC, OR a covering test exists but its mapped TC still describes OLD/superseded behavior (stale TC โ covered). Both the missing-TC and the stale-but-covered case are SPEC-GAP. Fix via $spec [mode=tests] UPDATE (and $spec when business rules changed)
FAIL:
Changed handler/command/entity rule with zero covering test
Unit test substituted where an integration test is feasible, with no justification
Test exists but does not assert the changed outcome (stale coverage counted as coverage)
New/changed behavior absent from spec docs, or TC describing superseded behavior
A COVERED row marked COVERED without reading its mapped ยง8 TC โ TC existence assumed instead of its CURRENT-behavior correctness verified (a stale-but-covered TC silently passes as covered)
Explicit user waiver (recorded verbatim in the report with the user's reason) is the ONLY alternative to closing a GAP.
Review Protocol (9 Phases)
Use task tracking for EACH phase before starting.
Phase 1 โ Collect: Split the change set: production files (Gate 7 coverage targets) vs test files. Categorize test files: new (full review), modified (changed methods only), new projects (infra + samples). Categorize production files: behavior-changing vs excluded (with reason).
Phase 2 โ Gate Review: Per test file, apply Gates 1-6. Apply Gate 7 once across the change set and produce the Coverage Mapping Table. Record per-file verdict table:
Gate
Verdict
Evidence
1. Assertion Value
PASS/FAIL
{file:line}
2. Data State
PASS/FAIL
{file:line}
3. Repeatability
PASS/FAIL
{file:line}
4. Domain Logic
PASS/FAIL
{file:line}
5. Traceability
PASS/WARN
{file:line}
6. Three-Way Sync
PASS/WARN/FAIL/ESCALATE
{file:line}
7. Change Coverage (per change set)
COVERED/COVERED-UNIT/GAP/SPEC-GAP
{coverage mapping table}
Phase 3 โ Spec Cross-Check + Three-Way Diff: Two directions โ from tests AND from changed code.
For each TC ID in code:
Verify TC entry exists in both docs/specs/ (Section 8) and docs/specs/
Read what TC describes in each doc
Read what implementation code actually does
Read what test asserts
Classify conflict pattern (Gate 6 table) and record action
Flag gaps both directions: TC in code but not in docs, or "Implemented" TC in docs but no test found
For each behavior-changing production file in the review target (reverse direction โ spec-driven development check):
Verify a TC exists describing the changed behavior AND read it to confirm it describes the CURRENT behavior โ run this even when a covering test was already found and the row is otherwise COVERED. If the TC still describes the OLD behavior, downgrade the row from COVERED to SPEC-GAP and flag it stale (route to $spec [mode=tests] UPDATE). Finding a covering test never excuses re-checking the TC's correctness.
New behavior with no TC anywhere โ SPEC-GAP finding (Gate 7); recommend $spec [mode=tests] (and $spec when business rules changed)
Phase 4 โ Initial Report: Write to plans/reports/integration-test-review-{date}-{slug}.md
Phase 5 โ Fix All Issues (MANDATORY): Fix every CRITICAL and HIGH issue. MEDIUM: fix if straightforward, document as tech debt otherwise.
Prioritize: CRITICAL โ HIGH โ MEDIUM
Per fix: read handler source, understand domain logic, write/fix assertion
Gate 7 GAP fixes: WRITE the missing test โ integration test first (route through $integration-test patterns); unit test only with recorded justification. SPEC-GAP fixes: run $spec [mode=tests] UPDATE to add/correct the TC before or alongside writing the test
NEVER weaken assertions to make tests pass โ fix root cause (timing, data, setup) instead
Re-read changed files to verify fix correctness
Record each fix with file:line under ## Fixes Applied
Phase 6 โ Validated Fix + Full Re-Review (MANDATORY when fixes are applied):
Do not spawn a fresh reviewer to re-review the same findings before validation/fix. After Phase 5 applies validated fixes, run a full fresh review over the current test scope. When that review uses sub-agents, spawn fresh integration-tester sub-agents (parallel by module for 10+ files; single agent otherwise) using canonical Agent template from SYNC:review-protocol-injection. Each sub-agent re-reads ALL target test files from scratch with ZERO memory of Phase 2/5. When constructing Agent call prompt:
Copy Agent call shape from SYNC:review-protocol-injection template verbatim
Set agent_type: "integration-tester"
Embed full verbatim body of 9 SYNC blocks (all present inline in this skill file): SYNC:evidence-based-reasoning, SYNC:bug-detection, SYNC:design-patterns-quality, SYNC:logic-and-intention-review, SYNC:test-spec-verification, SYNC:fix-layer-accountability, SYNC:rationalization-prevention, SYNC:graph-assisted-investigation, SYNC:understand-code-first
Task field: "Run a full fresh integration-test review pass over {file-list} after validated fixes were applied. Review against 7 quality gates: assertion value, data state, infinite repeatability, domain logic, test-spec traceability, three-way sync, change coverage. Read handler source AND feature docs before judging assertions. Flag smoke-only, existence-only, dead assertions, and repository-created invalid test data as FAIL. Gate 7: map every behavior-changing production file in {changed-production-file-list} to a covering test (integration-first; unit fallback requires justification) AND a spec TC โ uncovered behavior is a HIGH finding minimum, missing/stale TC is a SPEC-GAP finding. Source-of-truth hierarchy: feature docs > test-spec docs > implementation code > test code. Classify every disagreement as: wrong test, code bug, stale docs, or escalate (three-way conflict)."
Target Files: explicit file list (never pass inline contents)
Reference Docs: include docs/project-reference/integration-test-reference.md
Integrate findings as ## Re-Review {N} Findings โ DO NOT filter or override
If new CRITICAL/HIGH: validate the new finding set before any additional fixes
Repeat only after another fix cycle: restart the full review again after validated fixes are applied; if the same blocker repeats across 3 full invocations with no progress, escalate via a direct user question
Exit criteria: A complete full review returns 0 CRITICAL and 0 HIGH issues
Phase 7 โ Build & Run Tests (MANDATORY): Build and run ALL changed/reviewed test files.
Build test project
Run changed tests (filter by reviewed test classes)
NEVER mark review complete until all tests pass โ unverified reviews have zero value
Record results under ## Test Execution Results
Phase 8 โ Failure Investigation (if Phase 7 fails): Investigate systematically (classify โ root-cause โ fix plan), never just retry.
Classify failure: Test bug (assertion/setup wrong) vs Service bug (handler broken) vs Environment (service not running, DB timeout)
Fix plan per failure: failing test (file:line, TC-ID), error summary, root cause + confidence %, proposed fix
Apply and rerun โ loop until pass or environment blockers identified
Environment blockers: Document as BLOCKED โ requires running system; do NOT mark as test failures
Append under ## Failure Investigation
10+ files: Parallel sub-agents grouped by module. Each gets file list + 7 gates + handler paths + feature doc paths + the changed-production-file list for its module (Gate 7). Consolidate into single report โ the orchestrator merges per-module coverage tables into ONE Coverage Mapping Table covering the whole change set.
Common Anti-Patterns
Anti-Pattern
Why It's Bad
Smoke-only (no-exception alone)
Proves no crash, not correctness
Existence-only (not-null)
Proves data exists, not handler set it correctly
Dead assertion (count >= 0, always true)
Tests nothing
Framework testing (assert auto-set fields)
Tests framework, not handler
Copy-paste assertions (wrong entity fields)
Assertions don't match handler
Hardcoded ID (Id = "test-001")
Fails on second run
Cleanup dependency (finally { Delete(); })
Fragile, hides pollution
Order dependency (test B needs A first)
Parallel execution breaks
Repository data hacks (direct create/update bypassing use cases)
Leaves impossible state and hides real workflow bugs
Execute $integration-test-review directly โ run standalone
Phase 9: Why-Review Self-Validation Gate (MANDATORY when findings exist)
Purpose: Adversarial validation of own findings BEFORE handoff. Catches over-flagged Highs, false positives, and severity inflation at the source rather than letting them propagate downstream.
Trigger: Any finding produced (Critical, High, Medium, OR Low). Skip ONLY when the report's verdict is unconditional PASS with literally zero findings.
Protocol:
Read own finalized report from plans/reports/{skill}-{date}-{slug}.md
Invoke $why-review skill with arg: validate findings in plans/reports/{skill}-{date}-{slug}.md โ verify each finding has file:line proof, steel-man each rejected interpretation, and stress-test severity classifications
Read the validation verdict path returned by why-review, expected as plans/reports/why-review-validate-{date}.md
If why-review demotes/removes any finding: UPDATE own finalized report with revised severities, remove false positives, and add a ## Why-Review Validation Notes section citing what changed and why
If why-review confirms all findings: Append ## Why-Review Validation line to own report stating "All N findings re-validated against actual code; no severity changes."
Skip conditions (record explicit reason if skipping):
Verdict is unconditional PASS with zero findings โ log "Skipped โ no findings to validate"
Why-review skill itself is the active context (avoid recursion)
Why this exists: AI sub-agent reports inherit confirmation bias โ the orchestrator absorbs severity claims as ground truth. The 2026-05-09 review incident produced 5 Highs; adversarial validation demoted 3 of them. Codify this as standard practice.
Next Steps
MANDATORY โ NO EXCEPTIONS after completing, MUST use a direct user question:
"$integration-test-verify (Recommended)" โ Run integration tests to verify all pass
"$workflow-review-changes" โ Review all changes before committing
If Gate 5 fails (orphaned test) โ run $spec [mode=tests] UPDATE
$spec-index
Spec authority โ Gate 6 compares test code vs spec bundle
If Gate 6 finds conflict: spec is authority
$spec
Business doc โ Gate 6 compares tests vs feature doc business rules
If Gate 6 finds conflict: check spec vs spec-index alignment first
$docs-update
Orchestrator โ includes spec [mode=sync]
Call when Gate 6 reveals doc staleness
Standalone Chain
When called outside a workflow, follow this chain after running integration-test-review.
integration-test-review (you are here)
โ
โโ SCOPE: the full change set โ changed production code AND changed test files
โ Tests may NOT exist yet for changed code โ that is a Gate 7 finding, not an exit condition
โ
โโ Gate 1-5 findings โ fix tests (re-run integration-test if test code needs regeneration)
โ
โโ Gate 7 (Change Coverage) gap resolution:
โ โ
โ โโ GAP (changed behavior, no covering test):
โ โ โ Write the missing test โ integration-first via $integration-test
โ โ โ Unit test fallback ONLY when integration infeasible โ record justification
โ โ โ User waiver (verbatim, with reason) is the only alternative
โ โ
โ โโ SPEC-GAP (changed behavior, no/stale TC in spec docs):
โ โ $spec [mode=tests] UPDATE to add or correct the TC
โ โ $spec [update] when business rules changed
โ โ Then link the new/updated TC to the covering test (Gate 5)
โ
โโ Gate 6 (Three-Way Sync) conflict resolution:
โ โ
โ โโ Test code โ spec (feature doc says behavior A, test asserts behavior B):
โ โ โ Determine: spec authoritative or test authoritative?
โ โ โ If SPEC is correct: fix test โ re-run $integration-test
โ โ โ If TEST reflects correct new behavior (spec stale): $spec [update] โ $spec [mode=tests] [UPDATE] โ update test
โ โ
โ โโ Test code โ implementation (test asserts X, code does Y):
โ โ โ If CODE is correct: fix test โ $spec [mode=tests] UPDATE (update TC to match code's correct behavior)
โ โ โ If TEST is correct (code bug): do NOT update test โ fix code โ $prove-fix โ re-run tests
โ โ
โ โโ Derived index โ Feature Spec (the bucket INDEX.md / ERD disagrees with the canonical ยง1-8):
โ โ The Feature Spec is canonical; the index is regenerable, never authoritative
โ โ Run $spec-index to re-derive the index from the specs
โ โ Do NOT self-resolve โ escalate to user if ambiguous
โ
โโ [REQUIRED] โ $integration-test-verify
โ After all fixes, run actual tests to confirm all gates pass.
โ
โโ [REQUIRED] โ $spec [mode=sync]
โ If TCs were updated (Gate 5/6 fix), reconcile ยง8 TCs โ integration test code.
โ
โโ [RECOMMENDED] โ $docs-update
If Gate 6 revealed doc staleness, $docs-update runs full chain to update all layers.
[IMPORTANT] Use task tracking to break ALL work into small tasks BEFORE starting.
A test that cannot fail is not a test โ it is decoration. Every test MUST earn existence by proving it would FAIL if the protected business rule/invariant changed or the bug it guards were reintroduced.
Every finding requires file:line proof with confidence >80%.
Fresh Context Re-Review โ 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 โ do NOT spawn a confirmation sub-agent. A review round that finds issues triggers: validate findings โ fix โ 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 โ use integration-tester agent_type (integration-test reviews ALWAYS spawn integration-tester, NOT code-reviewer)
Inject ALL required review protocols VERBATIM into the prompt โ 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 โ 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 โ every fix invalidates the prior verdict
NEVER reuse a sub-agent across rounds โ 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 via a direct user question
Track iteration count and repeated blockers in conversation context (session-scoped, no persistent files)
Sub-Agent Type Override
MANDATORY: Integration-test reviews spawn the integration-tester sub-agent, NOT code-reviewer.
Keep agent_type: "integration-tester" from the canonical template below; NEVER revert to code-reviewer.
Rationale:integration-tester specializes in test-spec generation, TC traceability, CQRS test patterns, async-polling / eventual-consistency assertion correctness, and cross-service integration context โ areas code-reviewer does not cover at depth.
Review Protocol Injection โ 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.
code-reviewer โ for general code-quality reviews only (NOT integration tests)
Canonical Agent Call Template (Copy Verbatim)
spawn_agent({
description: "Fresh Round {N} review",
agent_type: "integration-tester",
prompt: `
## Task
{review-specific task โ 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 โ these are non-negotiable)
### Spec โ Tests โ Code Triangulation
DO THIS FIRST โ before any per-protocol check below. The review target is the WHOLE PACKAGE, not the diff alone: load the behavior's spec (ยง3 ACs / ยง4 BRs / ยง8 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 โ every disagreement is a finding; classify which face is wrong:
- code vs spec: behavior the code does that no ยง3/ยง4/ยง8 rule describes โ CODE-EXTRA or SPEC-STALE; a [HARD] ยง4 rule or ยง5 invariant with no enforcing code path โ CODE-WRONG.
- tests vs spec: a ยง8 TC with no test, or a test asserting behavior no TC/rule names โ TEST-GAP or SPEC-SILENT.
- tests vs code: a changed code path with no covering test โ TEST-GAP; a test that still passes against a deliberately broken invariant โ 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 ยง3/ยง4/ยง8 AND guarded with a test โ the enrichment loop, never a silent pass.
4. Only after the three faces agree โ or every disagreement is logged as a finding โ 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 โ 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) โ FAIL | HIGH (incorrect behavior) โ FAIL | MEDIUM (edge case) โ WARN | LOW (defensive) โ 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 โ 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 โ 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 โ verify corresponding authorization and data-state test cases exist.
8. If no specs exist for a changed path โ 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 ร pre-fix ร post-fix ร delta table BEFORE writing verdict.
- Minimum 3 rows; include at least one row OUTSIDE the original bug report.
- Any "REGRESSION" delta โ 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 โ 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 โ Map the complete path from data origin to crash site across ALL layers (storage โ backend โ API โ frontend โ UI). Identify where bad state ENTERS, not where it CRASHES.
2. Identify the invariant owner โ 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 โ If fix requires touching 3+ files with defensive checks, you are at the wrong layer โ go lower.
4. Verify no bypass paths โ 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 โ 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 โ 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" โ Simple + wrong assumptions = wasted time. Plan anyway.
- "I'll test after" โ RED before GREEN. Write/verify test first.
- "Already searched" โ Show grep evidence with file:line. No proof = no search.
- "Just do it" โ Still need task tracking. Skip depth, never skip tracking.
- "Just a small fix" โ Small fix in wrong location cascades. Verify file:line first.
- "Code is self-explanatory" โ Future readers need evidence trail. Document anyway.
- "Combine steps to save time" โ 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 โ trace --direction both reveals full system flow โ 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) โ cite file:line evidence.
2. Read existing files in target area โ 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 โ 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 โ never work from memory alone.
7. NEVER invent new patterns when existing ones work โ 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)
- docs/project-reference/code-review-rules.md
- {skill-specific reference docs โ 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 โ 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 integration-tester agent_type โ integration-test reviews ALWAYS use integration-tester, never code-reviewer
DO NOT paraphrase, summarize, or skip any protocol section
DO NOT pass file contents inline โ 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 โ the bodies are already embedded above
DO NOT introduce placeholder markers for the protocols โ they must stay literally expanded
Critical Thinking Mindset โ Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80% to act.
Anti-hallucination: Never present guess as fact โ cite sources for every claim, admit uncertainty freely, self-check output for errors, cross-reference independently, stay skeptical of own confidence โ certainty without evidence root of all hallucination.
Evidence-Based Reasoning โ 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 โ output: "Insufficient evidence. Verified: [...]. Not verified: [...]."
Validated-Finding Fix + Full Re-Review Loop โ Re-review is triggered by a validated finding fix cycle, not by a round number. Review purpose: review โ validate findings โ fix validated findings โ full re-review until a complete review pass finds no issues. A clean review ENDS the loop โ no further rounds required.
No issues found (PASS, zero findings) โ review ENDS. Do NOT spawn a fresh sub-agent for confirmation.
Issues found (FAIL, or any non-zero findings) โ 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 โ 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 โ END; issues โ validate findings โ fix โ restart from the first review phase. Continue until a complete review pass finds zero issues. If the same validated finding repeats for 3 full invocations with no progress, or a fix requires product/owner input, escalate via a direct user question.
Rules:
A clean Round 1 ENDS the review โ no mandatory Round 2
NEVER fix unvalidated findings; validate first using the caller's validation gate
NEVER skip the full re-review after a fix cycle (every fix invalidates the prior verdict)
NEVER reuse a sub-agent across rounds โ every iteration that uses sub-agents spawns NEW Agent calls
Main agent READS sub-agent reports but MUST NOT filter, reinterpret, or override findings
No arbitrary sub-agent-round cap replaces the clean-review requirement; use the 3 repeated-no-progress blocker rule only to avoid infinite spinning
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 Nโฅ2 that was executed.
Infinitely Repeatable Tests โ Tests MUST run N times without failure. Like manual QC โ run the suite 100 times, each run just adds more data. Verification is only PASS after the relevant suite/project passes 3 consecutive runs without database reset.
Unique data per run: Use the project's unique ID generator for ALL entity IDs created in tests. NEVER hardcode IDs.
Additive only: Tests create data, never delete/reset. Prior test runs MUST NOT interfere with current run.
No schema rollback dependency: Tests work with current schema only. Never rely on schema rollback or migration reversals.
Idempotent seeders: Fixture-level seeders use create-if-missing pattern (check existence before insert). Test-level data uses unique IDs per execution.
No cleanup required: No teardown, no database reset between runs. Each test is isolated by unique seed data, not by cleanup.
Unique names/codes: When entities require unique names/codes, append a unique suffix using the project's ID generator.
Migration code excluded: Do not write tests for migration code. Schema/data migrations are one-time execution paths, not core application logic.
Source/test drift check. For coding, fix, debug, investigation, test, or review work: when source behavior changes, inspect affected unit/integration/E2E tests and decide from evidence whether tests should change to match intended behavior or the source change is an unintended bug to fix. Do not write tests for migration code; schema/data migrations are one-time execution paths, not core application logic.
AI Mistake Prevention โ 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 โ ask WHY before changing. Before changing a constant, limit, flag, wording, or pattern, read nearby context and history.
Surface ambiguity before acting โ 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.
Nested Task Expansion Contract โ 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 โ 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 โ 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 every required doc. 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 $project-init or the narrow lower-level route ($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.
Before target work, state: Reference docs read: ... | Not applicable: ....
Task Tracking & External Report Persistence โ 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.
Systematic Review Batching (map-reduce) โ When a changeset is large, do NOT review files one-by-one. Partition into size-capped batches, fire one specialized sub-agent per batch in parallel, then reduce. This bounds EVERY context โ each batch agent AND the orchestrator โ so coverage stays complete as file count grows.
Trigger ladder (one ordered escalation โ not competing thresholds):
categories > 6 OR files > 40 โ additionally insert the hierarchical synthesis tier (below). Everything from rung 2 still applies.
Step 1 โ Categorize. Group changed files into logical categories derived from the project's actual structure (not forced). Category is the concern axis; orient with these examples, derive what fits the repository:
Category Type
Example Groupings
Agent/Tooling
AI scripts, hooks, skill definitions, workflow configs, linting rules
Service/handler/controller source (infer from project structure)
Frontend logic
UI component/state/API source (infer from project structure)
Data/Schema
Migrations, schema files, seed data
Tests
Unit, integration, E2E test files
Infrastructure
Docker, k8s, CI/CD, cloud manifests
Step 2 โ Size-capped batches. One sub-agent per batch of โค8 files OR โค2000 diff-lines, whichever hits first. Category stays the concern axis, but any category exceeding a cap splits into multiple size-capped batches (30 backend files โ 4 batches). Size caps โ not category caps โ make "many files" safe: a category cap alone lets one giant category blow a single agent's context.
Step 2a โ Sub-agent type per batch (match the batch's dominant concern):
Docs, plans, specs, configs, infra โ general-purpose
Each batch sub-agent receives: its full file list; SYNC:category-review-thinking as its primary thinking model โ derive each category's concerns from first principles, NOT a fixed checklist (if the consuming skill does not carry that block, apply category-first thinking directly); project reference docs relevant to its concern (discover via *patterns*, *conventions*, *style-guide*); cross-reference verification instructions (counts, tables, links). All batch agents run in parallel and write findings to plans/reports/ (per SYNC:task-tracking-external-report); reducers read from disk, never from memory.
Step 3 โ Reduce.
Flat reduction (rung 2, โค6 categories AND โค40 files): the orchestrator collects each batch report, cross-references counts/tables/contracts ACROSS batches, detects gaps visible only across categories (feature in code but missing from docs; new API endpoint with no client call), and consolidates into one categorized holistic report.
Hierarchical reduction (rung 3, > 6 categories OR > 40 files): insert a mid-tier โ each concern gets ONE synthesizer agent that reads only its own batch reports and emits a single concern-synthesis. The orchestrator reads the concern-syntheses (~5), never the raw batch reports โ keeping the reducer's context O(#concerns), not O(#files).
Cross-concern interaction pass (mandatory at rung 3 โ closes the synthesis-tier blind spot): concern-siloed synthesis can drop an interaction spanning two concerns AND two batches (tainted source in data-layer/batch 7 โ sink in api/batch 3). So: (a) each concern-synthesizer MUST emit an explicit "cross-concern interaction candidates" list โ entities/symbols/contracts it touched that plausibly bind to another concern (shared DTOs, event names, table/collection names, exported symbols); (b) the orchestrator MUST run the Step-3 cross-reference/gap step over those candidate lists across all concern-syntheses, not only within a batch, before concluding. Without this pass the tier trades completeness for context-bounding on exactly the large diffs it targets.
Step 4 โ Holistic assessment. With all findings combined, judge: overall coherence as a unified intent; cross-category sync (docs match code? contracts match callers?); risk areas where categories interact; missing doc/spec updates for changed artifacts.
No silent truncation. If any cap forces sampling or a batch is dropped for budget, ANNOUNCE the dropped/sampled scope explicitly โ bounded coverage must never read as complete coverage.
Severity Rubric โ 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
