メニュー
Run a repeatable code review audit on migration changes. Inventories files by architectural layer, reviews each layer against #1050/#1066 principles, and files beads issues for findings. Re-run after remediation batches to track progress.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Start an isolated PostgreSQL container for integration testing. Each worktree gets its own container on a unique port — no mutex, no conflicts. Use this before running integration tests in a worktree agent.
Two-pass BDD step assertion completeness inspector. Pass 1 (Sonnet): triage all Then steps — FLAG or PASS. Pass 2 (Opus): deep trace flagged steps with full production context, producing architectural judgment on what the correct assertion should be. Use after writing or modifying BDD step definitions to catch assertion mismatches (steps that claim to verify X but actually only check existence).
Derive obligation tests using the test harness. First checks for harness availability, then generates compact tests using domain-specific test environments. Hard gate: no harness -> no tests (stops immediately). Replaces obligation-test for harness-enabled domains.
Create structural guard tests that enforce architecture principles on every `make quality` run. Guards are AST-scanning tests that prevent categories of violations automatically. Available guards: schema-inheritance, boundary-completeness, query-type-safety, no-error-dicts.
Derive integration tests from existing unit test xfails and UNSPECIFIED stubs. Uses integration-from-stub formula: catalog → review → triage → architect → write-integration → fix-green → reconcile-xfail → verify → commit. Each stub's expected behavior is the core invariant. Architect atom reads 7 architecture docs and cross-references against CRIT-1..CRIT-11 findings.
Map the complete test surface for domain entities by cross-referencing test-obligations with existing tests. Produces one canonical test module per entity with real tests + skip stubs for gaps. Every obligation maps to exactly one test. Run this before /remediate.
| name | audit |
| description | Run a repeatable code review audit on migration changes. Inventories files by architectural layer, reviews each layer against #1050/#1066 principles, and files beads issues for findings. Re-run after remediation batches to track progress. |
| args | ["audit-target"] |
Repeatable code review workflow for auditing migration changes. Produces structured review documents and beads issues for every finding.
/audit [audit-target]
The audit target (optional, defaults to "full"):
KonstantinMirin/adcp-v3-upgrade — review all changes on branchmain..HEAD — review commits in rangedocs/code-reviews/python3 .claude/scripts/cook_formula.py \
--formula .claude/formulas/migration-audit.yaml \
--var "AUDIT_TARGET={all_args}" \
--epic-title "Audit: {all_args}"
If no target specified, use "full":
python3 .claude/scripts/cook_formula.py \
--formula .claude/formulas/migration-audit.yaml \
--var "AUDIT_TARGET=full" \
--epic-title "Audit: full codebase"
Dry run first (recommended):
python3 .claude/scripts/cook_formula.py \
--formula .claude/formulas/migration-audit.yaml \
--var "AUDIT_TARGET={all_args}" \
--epic-title "Audit: {all_args}" \
--dry-run
bd ready → bd show <atom-id> → execute → bd close <atom-id> → repeat
Linear pipeline: inventory-changes → review-per-layer → consolidate → file-issues → commit-report.
Audit report in docs/code-reviews/, beads issues filed for all findings.
Each layer has a specific checklist (see formula for full details):
| Layer | Key Checks |
|---|---|
| Schema | Correct base class, no field duplication, exclude=True on internals |
| Business | No ToolError, no transport imports, ResolvedIdentity, no error dicts |
| Boundary | All _impl params exposed, version compat at boundary only |
| Transport | Shared _impl pattern, no business logic duplication |
| Database | SQLAlchemy 2.0, JSONType, correct column types |
| Adapter | No protocol code, proper error propagation |
The audit is designed to be re-run:
/audit main..HEAD # after first batch
/audit main..HEAD # after second batch (findings should decrease)
Compare reports across runs to track progress.
/surface — Create entity test suites for coverage gaps found in audit/guard — Structural guards enforce the principles audit checks for/remediate — Fill test stubs to fix the issues audit finds/mol-execute — Execute individual beads tasks for specific findings