Quick Code Review

Fast, cheap review using 4 parallel haiku agents. For comprehensive PR review with 6 specialized agents, use /pr-review instead.

When This Skill Applies

• User asks for a quick review
• User says "/pr-review-quick"
• Fast feedback before committing

Workflow

Step 1: Get the Diff

# Get changed files
git diff main...HEAD --name-only 2>/dev/null || git diff HEAD~3 --name-only

# Get the full diff
git diff main...HEAD 2>/dev/null || git diff HEAD~3

Step 2: Launch Parallel Review Agents

CRITICAL: Launch ALL of these agents in a SINGLE message with multiple Task tool calls.

Use model: haiku for each agent to keep costs low.

Every agent MUST receive this instruction:

IMPORTANT: Only report issues on lines that were ADDED or MODIFIED in this diff (+ lines).
Do NOT report pre-existing issues in unchanged code.

Task (model: haiku): "SECURITY REVIEW

CRITICAL: Only flag issues on ADDED/MODIFIED lines (+ lines in diff). Ignore pre-existing code.

Check this diff for security issues:
- SQL injection
- XSS vulnerabilities
- Hardcoded secrets/credentials
- Missing input validation

Diff:
[paste diff]

Output format (only for issues with confidence ≥ 75%):
[HIGH/MEDIUM/LOW] file:line - description

If no high-confidence issues: 'No security issues found'"

Task (model: haiku): "BUG REVIEW

CRITICAL: Only flag issues on ADDED/MODIFIED lines (+ lines in diff). Ignore pre-existing code.

Check this diff for bugs and logic errors:
- Off-by-one errors
- Null/undefined handling
- Missing edge cases
- Race conditions
- Incorrect conditionals

Diff:
[paste diff]

Output format (only for issues with confidence ≥ 75%):
[HIGH/MEDIUM/LOW] file:line - description

If no high-confidence issues: 'No bugs found'"

Task (model: haiku): "PERFORMANCE REVIEW

CRITICAL: Only flag issues on ADDED/MODIFIED lines (+ lines in diff). Ignore pre-existing code.

Check this diff for performance issues:
- N+1 queries
- Unnecessary loops/iterations
- Missing memoization
- Large bundle imports
- Memory leaks

Diff:
[paste diff]

Output format (only for issues with confidence ≥ 75%):
[HIGH/MEDIUM/LOW] file:line - description

If no high-confidence issues: 'No performance issues found'"

Task (model: haiku): "CODE QUALITY REVIEW

CRITICAL: Only flag issues on ADDED/MODIFIED lines (+ lines in diff). Ignore pre-existing code.

Check this diff for code quality:
- Unclear naming
- Overly complex logic
- Code duplication
- Missing error handling
- Inconsistent patterns

Diff:
[paste diff]

Output format (only for issues with confidence ≥ 75%):
[HIGH/MEDIUM/LOW] file:line - description

If no high-confidence issues: 'No quality issues found'"

Step 3: Synthesize Results

Collect all agent responses and combine into a single report.

Filter out:

• Issues on lines not in the diff
• Low-confidence findings (nitpicks, maybes)
• Things a linter would catch

## Review Summary

### Security
[agent 1 findings, or "No issues"]

### Bugs
[agent 2 findings, or "No issues"]

### Performance
[agent 3 findings, or "No issues"]

### Code Quality
[agent 4 findings, or "No issues"]

---
Files reviewed: [count]
Issues found: [HIGH: n, MEDIUM: n, LOW: n]

Output Format

If issues found:

[HIGH] path/to/file.ts:42 - SQL injection vulnerability in user query
[MEDIUM] path/to/other.ts:15 - Missing null check on optional param

If no issues: "No issues found" with a brief note on what was reviewed.