audit-code

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.
disable-model-invocation	true

Audit Code

HARD GATE — 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.

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 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 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)

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

Agent Readability (Akita's Lens)

• Functions are small enough to fit in a standard context window (4–20 lines)
• Names are unique and specific enough to be grep-able (grep returns < 5 hits)
• Types are explicit (no any, no inferred return types for public APIs)
• Code avoids deep nesting (max 2 levels) and uses early returns

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 && echo "OK: audit-code dependencies present" || echo "FAIL"

Handoff

Gate: READY -> next: commit-message Writes: state.yaml handoff.next_skill = commit-message

---

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.