audit-code · git:20261025.6136322 · 2026-10-25 · sha256 703c37e96526d45a
audit-code git:20261025.6136322A
Immutable. This exact content never changes and is served at /api/v1/blob/703c37e96526d45a.
No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_gdfkmbblmpapjr3a/703c37e96526d45a68de4485c1cefda563918315b569e424541eb31380e5f127.json
Then /plugin install audit-code@markdownregistry-gdfkmbblmpapjr3a in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.
--- name: audit-code model: haiku description: "Self-review checklist for the coding agent to run before dispatching a reviewer. Checks CONVENTIONS.md compliance, Boy Scout Rule, test coverage, types, and SOLID. Produces a pass/fail checklist. Use before request-review, before committing, or when user asks for a code quality check." --- # Audit Code > **HARD GATE** — Audit must check for: bugs (correctness), security, performance, and clarity. Do NOT skip security review if the code touches user data, auth, or external APIs. ## Target (argument) If an argument was passed after `/audit` (e.g. `/audit abc1234` or `/audit HEAD~3..HEAD`), use it as the diff target: ```bash git diff <arg> # range or two-ref form: abc..def, HEAD~3..HEAD git show <arg> # single commit: abc1234 ``` If no argument was given, review uncommitted working-tree changes (`git diff` + `git diff --staged`). Run this self-review before asking anyone else to look at the code. The goal is to catch everything that is clearly wrong or missing — so the reviewer can focus on design and architecture, not hygiene. **Distinct from `request-review`:** This is the coding agent checking its own work. No second agent is involved. Run this first; run `request-review` after this passes. ## Look-here-first (churn heuristic) Before the checklist, rank changed files by git churn and review **high-churn hotspots first** — they carry the most latent risk regardless of diff size. ```bash bash scripts/bp-churn-rank.sh --since 90.days --limit 15 ``` Apply the full checklist to churn-ranked files in descending order. Files with zero recent commits but large diffs still get reviewed; churn only sets priority, not scope. ## Modes - Default: full checklist - --quick: Run only Supply Chain and Test Coverage. Use for changes under 50 LOC. - --gate: Non-interactive mode for automated CI gating (used by build-epic step 6). Exit with non-zero status code (`exit 1`) on ANY checklist failure; `exit 0` only if ALL items pass. Produces a compact pass/fail summary to stderr. On failure, list every ✗ item with reason. - --parallel: Run checklist sections in **isolated git worktrees** (e45s18) so concurrent checks cannot corrupt each other's working tree: ```bash bash scripts/lib/parallel-review-worktrees.sh audit-code ``` ## Checklist ### Supply Chain & Security - [ ] slopcheck run for new dependencies; packages tagged in plan-work: `[OK]`, `[SUS]`, or `[SLOP]` - [ ] No `[SLOP]` packages without documented human approval - [ ] No secrets in diff (`sk-`, `ghp_`, `AKIA`, `.env` values) — see `guard-git` patterns - [ ] OWASP Top 10 spot-check: injection, broken auth, sensitive data exposure, misconfiguration (see `docs/references/security-threats.md`) - [ ] Security: diff scanned — no unaddressed HIGH findings (or deviations documented in `specs/security/EXCEPTIONS.md`) ### Provenance & Metadata - [ ] New plan artefacts include `type:` and `context:` metadata - [ ] Implementation steps reference ADR or commit SHA where decisions were made ### Law of Demeter - [ ] No method chains through unrelated objects (e.g. `a.getB().getC().doX()`) - [ ] Collaborators talk to immediate neighbors only; law violations need explicit justification ### CONVENTIONS.md Compliance - [ ] All output files are in `specs/` (no docs written to project root) - [ ] No `gh issue create` calls anywhere in new/modified skills or scripts - [ ] `gh` used only for PRs and repo clone operations - [ ] No GitHub REST API called directly (no curl/fetch to api.github.com) ### Scope - [ ] Changes are limited to what was asked — nothing extra refactored or reorganized - [ ] No speculative features added - [ ] No files touched outside the stated scope - [ ] **Discovered defects:** Reproducible gate failures (Preflight, CI, golden suite) require fix-or-log — `quick-fix` or `fix-bug` — even when "outside" the story scope. Scope-minimization does not waive Always Green. - [ ] Boy Scout Rule applies to files opened to fix a gate failure; it does not excuse skipping red Preflight ### Boy Scout Rule - [ ] Every file I touched is cleaner than when I found it - [ ] No dead code left behind - [ ] No commented-out code blocks ### Types and Safety - [ ] No `any` types introduced (TypeScript) or untyped public functions (Python/Go/etc.) - [ ] No `@ts-ignore` or `// eslint-disable` added - [ ] No `as unknown as X` casts that bypass type safety ### Test Coverage - [ ] Every new function has at least one test - [ ] Every bug fix has a regression test - [ ] Tests verify behavior through public interfaces (not implementation details) - [ ] Tests are F.I.R.S.T compliant (per CONVENTIONS.md §Tests; use `enforce-first` if unsure) ### SOLID and Heuristics - [ ] Single Responsibility: no function or module doing two unrelated things - [ ] Open/Closed: extended through interfaces, not by modifying stable code - [ ] Dependency Inversion: dependencies injected, not imported globally where avoidable - [ ] **Chapter 17 Heuristics**: Code is free of smells documented in `audit-code/HEURISTICS.md` (G, N, C, T) ### Refactoring Smells (Fowler) Explicitly name any detected smells: Mysterious Name, Duplicated Code, Feature Envy, Data Clumps, Primitive Obsession, Message Chains, Middle Man. ### Code Style (CONVENTIONS.md) - [ ] Functions: 4–20 lines; split if longer - [ ] Functions: descend exactly one level of abstraction (The Stepdown Rule / G34) - [ ] Files: under 300 lines (ideally 200–300) - [ ] Names: specific and unique (grep returns < 5 hits for each name) - [ ] No duplication — shared logic extracted (DRY / G5) - [ ] Early returns over nested ifs; max 2 levels of indentation - [ ] Conditionals: expressed as positives (G29) - [ ] Comments explain WHY, not WHAT ### Red Flags Before reporting, name any rationalization you caught yourself making for skipping a checklist item. Silence is not acceptable — if you skipped an item, state the reason explicitly. ## Output Report the checklist with ✓ / ✗ per item. For each ✗, describe what needs to be fixed. If all items pass: suggest running `request-review` for an independent second opinion. If any items fail: fix them before proceeding. In `--gate` mode, print one summary line per checklist section (`PASS Supply Chain` / `FAIL Provenance (2 items)`). Exit `0` only if all PASS. Write full report to `specs/verifications/AUDIT-<epic>-<story>.md`. ## Verify → verify: `test -f CONVENTIONS.md && test -d skills/enforce-first && test -d skills/request-review` ## Handoff Gate: READY -> next: commit-message Writes: state.yaml handoff.next_skill = commit-message <!-- story: e01s02 --> <!-- story: e06s03 --> <!-- story: e07s01 --> --- # Clean Code Heuristics (Chapter 17) A summary of Robert C. Martin's catalogue of code smells and heuristics, used as the technical benchmark for `audit-code`. ## Comments (C) - **C1: Inappropriate Information**: Comments should only hold technical notes. Metadata (author, change history) belongs in Git. - **C2: Obsolete Comment**: Update or delete comments that are no longer accurate. - **C3: Redundant Comment**: Don't describe code that adequately describes itself (e.g., `i++; // increment i`). - **C4: Poorly Written Comment**: If you write a comment, spend time making it the best it can be. - **C5: Commented-Out Code**: Delete it. Git remembers it. ## Environment (E) - **E1: Build Requires More Than One Step**: Building should be a single trivial operation (e.g., `bash install.sh`). - **E2: Tests Require More Than One Step**: Running all tests should be one simple command (e.g., `npm test`). ## Functions (F) - **F1: Too Many Arguments**: 0 is ideal, 1-2 is fine, 3 requires special justification. Never > 3. - **F2: Output Arguments**: Avoid them. If a function changes state, it should change the state of its owning object. - **F3: Flag Arguments**: Boolean arguments are a smell that the function does > 1 thing. - **F4: Dead Function**: Discard methods that are never called. ## General (G) - **G1: Multiple Languages in One Source File**: Try to minimize the mixing of languages (e.g., HTML inside Java). - **G5: Duplication (DRY)**: **The root of all evil.** Every time you see duplication, it's a missed opportunity for abstraction. - **G6: Code at Wrong Level of Abstraction**: High-level concepts in base classes; low-level details in derivatives. - **G25: Replace Magic Numbers with Named Constants**: No "naked" numbers or strings. - **G28: Encapsulate Conditionals**: Prefer `if (shouldBePublished())` over complex boolean logic. - **G29: Avoid Negative Conditionals**: Prefer `if (buffer.shouldCompact())` over `if (!buffer.shouldNotCompact())`. - **G30: Functions Should Do One Thing**: If a function can be split into sections, it's doing too much. - **G31: Hidden Temporal Couplings**: If execution order matters, make the dependency explicit via arguments. - **G34: Functions Should Descend Only One Level of Abstraction**: The Stepdown Rule. ## Naming (N) - **N1: Choose Descriptive Names**: Names should reveal intent and be updated as code evolves. - **N4: Unambiguous Names**: Names should make the working of a function/variable clear. - **N7: Names Should Describe Side-Effects**: Describe everything the function is or does. ## Tests (T) - **T1: Insufficient Tests**: A test suite should test everything that could possibly break. - **T4: An Ignored Test Is a Question about an Ambiguity**: Document the reason for `@Ignore`. - **T5: Test Boundary Conditions**: Most bugs happen at the boundaries; test them exhaustively. - **T8: Test Coverage Patterns Can Be Revealing**: Analyze what code is *not* executed to find gaps. - **T9: Tests Should Be Fast**: Slow tests don't get run.