Code Review Standards
Purpose
This skill defines the structured, severity-tagged checklist the code-critic agent applies
when reviewing an implementation. Severity tagging makes review deterministic across
dispatches: PM and engineer both know exactly which findings block delivery and which are
advisory. Engineers may load it for self-review before requesting a critic pass.
The Severity-Tagged Checklist
CRITICAL (must fix — blocks delivery)
- No secrets, API keys, or credentials hardcoded
- No injection vectors (parameterized queries only; no unsafe shell/template interpolation)
- No arbitrary code execution paths (
eval, exec, unrestricted deserialization)
- Authentication/authorization not bypassable
- No data-loss or silent-corruption paths
HIGH (must fix — blocks delivery)
- Type hints / static types on public functions and classes; type checker passes clean
- Tests pass with zero failures; meaningful coverage on new code
- No bare
except, no swallowed errors, error cases handled explicitly
- No mutable default arguments, no unguarded global mutable state
- No synchronous I/O inside async functions; no N+1 query patterns
MEDIUM (flag, note in report, proceed)
- Functions reasonably small and single-purpose
- Hash maps used where nested loops would otherwise dominate complexity
- Async operations have explicit timeouts
- Docstrings on public methods; no stray
Any in production paths
LOW (note only)
- Formatter/linter clean (handled by tooling)
- Clear naming, no commented-out code, tidy imports
Verdict Protocol
First line of the critic response MUST be the verdict, followed by a finding table.
| Verdict |
Condition |
PM Action |
| APPROVE |
Zero CRITICAL, zero HIGH |
Proceed to security review |
| WARN |
Zero CRITICAL, one or more HIGH |
Proceed, but log finding table to handoff |
| BLOCK |
Any CRITICAL |
Halt, surface findings to user, await direction |
Finding table columns: Severity | File | Line | Issue | Required Fix.
80% Confidence Filter
A clean review is a valid review — do not manufacture findings. Only report issues you are
80% confident are real problems. Below that threshold, raise the concern as a question in
the summary paragraph rather than as a table row. Each table finding must be actionable: the
engineer should be able to fix it without asking for clarification.
1---2name: code-review-standards-33description: Severity-tagged code review rubric (CRITICAL/HIGH/MEDIUM/LOW) used by the code-critic agent to produce APPROVE/WARN/BLOCK verdicts with evidence-backed findings4version: 1.0.05category: collaboration6author: Claude MPM Team7license: MIT8progressive_disclosure:9 entry_point:10 summary: "Severity-tagged checklist for code review. CRITICAL/HIGH block delivery; MEDIUM is flagged; LOW is noted. Critic outputs APPROVE (zero CRITICAL/HIGH), WARN (HIGH but no CRITICAL), or BLOCK (any CRITICAL). 80% confidence filter — don't manufacture findings."11 when_to_use: "Loaded by the code-critic agent to apply structured review criteria. Also loaded by engineers for self-review before requesting a critic pass."12 quick_start: "Apply checklist top-to-bottom (CRITICAL first). For each finding cite file+line, explain the problem, give the fix. Filter to >80% confidence. Output a verdict line + finding table."13context_limit: 70014tags:15 - code-review16 - quality17 - rubric18 - standards19 - verdict20 - severity21requires_tools: []22metadata:23 source: https://skillmd.com/skills/bobmatnyc/code-review-standards-324---2526# Code Review Standards2728## Purpose2930This skill defines the structured, severity-tagged checklist the `code-critic` agent applies31when reviewing an implementation. Severity tagging makes review deterministic across32dispatches: PM and engineer both know exactly which findings block delivery and which are33advisory. Engineers may load it for self-review before requesting a critic pass.3435## The Severity-Tagged Checklist3637### CRITICAL (must fix — blocks delivery)38- No secrets, API keys, or credentials hardcoded39- No injection vectors (parameterized queries only; no unsafe shell/template interpolation)40- No arbitrary code execution paths (`eval`, `exec`, unrestricted deserialization)41- Authentication/authorization not bypassable42- No data-loss or silent-corruption paths4344### HIGH (must fix — blocks delivery)45- Type hints / static types on public functions and classes; type checker passes clean46- Tests pass with zero failures; meaningful coverage on new code47- No bare `except`, no swallowed errors, error cases handled explicitly48- No mutable default arguments, no unguarded global mutable state49- No synchronous I/O inside async functions; no N+1 query patterns5051### MEDIUM (flag, note in report, proceed)52- Functions reasonably small and single-purpose53- Hash maps used where nested loops would otherwise dominate complexity54- Async operations have explicit timeouts55- Docstrings on public methods; no stray `Any` in production paths5657### LOW (note only)58- Formatter/linter clean (handled by tooling)59- Clear naming, no commented-out code, tidy imports6061## Verdict Protocol6263First line of the critic response MUST be the verdict, followed by a finding table.6465| Verdict | Condition | PM Action |66|---------|-----------|-----------|67| APPROVE | Zero CRITICAL, zero HIGH | Proceed to security review |68| WARN | Zero CRITICAL, one or more HIGH | Proceed, but log finding table to handoff |69| BLOCK | Any CRITICAL | Halt, surface findings to user, await direction |7071Finding table columns: `Severity | File | Line | Issue | Required Fix`.7273## 80% Confidence Filter7475A clean review is a valid review — do not manufacture findings. Only report issues you are76>80% confident are real problems. Below that threshold, raise the concern as a question in77the summary paragraph rather than as a table row. Each table finding must be actionable: the78engineer should be able to fix it without asking for clarification.