/code-review — Comprehensive Code Review

Launch parallel review agents for a thorough code review with confidence-scored findings.

Process

1. Determine Review Target

• •No argument: Review all uncommitted changes (git diff + git diff --cached)
• •PR number: Fetch PR diff with gh pr diff <number>
• •File path: Review that specific file's recent changes

2. Analyze Changes

git diff --stat  # Overview of what changed
git diff         # Full diff

Identify:

• •Which crates are affected
• •Whether bench/template code changed
• •Whether auth/crypto/security code changed

3. Launch Agents in Parallel

Always launch:

• •code-reviewer (sonnet): Full 4-pass review with confidence scoring

Conditionally launch:

• •security-auditor (haiku): If changes touch auth, crypto, jwt, token, password, secret, registry, or middleware code
• •template-checker (haiku): If changes touch .hbs files, k6_gen.rs, command.rs, or any template data building code

4. Consolidate Results

Collect findings from all agents. Apply the 80+ confidence threshold:

• •Report findings with confidence >= 80
• •Suppress findings with confidence < 80 (mention count but not details)

5. Output

## Code Review Summary

### Changes Reviewed
- Files: N (list crates affected)
- Lines: +X / -Y

### Findings (confidence >= 80)
1. [CRITICAL] file:line — Description (confidence: 95)
2. [WARNING] file:line — Description (confidence: 82)

### Suppressed
- N low-confidence findings suppressed (use `/code-review --verbose` to see all)

### Security Audit
- (security-auditor results, if launched)

### Template Consistency
- (template-checker results, if launched)

### Recommendation: APPROVE | REQUEST_CHANGES | NEEDS_DISCUSSION

Rules

• •Always launch code-reviewer; other agents are conditional
• •The 80+ confidence threshold is the default; --verbose shows everything
• •Don't duplicate findings across agents — if security-auditor and code-reviewer both find the same issue, report it once