lucidlabs-hq/review

Review: Parallel Code Review Orchestration

Objective

Run specialized review agents in parallel against the codebase. Collect findings into a single structured report.

---

Process

1. Determine Scope

Parse the argument to determine what to review:

| Argument | Scope | |----------|-------| | /review or /review full | Run ALL 4 reviewers on frontend/ and convex/ | | /review security | Run security-reviewer only | | /review performance | Run performance-reviewer only | | /review simplicity | Run simplicity-reviewer only | | /review nextjs | Run nextjs-react-reviewer only | | /review <path> | Run ALL 4 reviewers scoped to the given path |

2. Launch Review Agents

Use the Task tool to launch specialized agents in parallel. Each agent corresponds to a file in .claude/agents/:

Agent Mapping:

| Review Type | Agent File | Focus | |-------------|-----------|-------| | Security | .claude/agents/security-reviewer.md | OWASP, auth, XSS, CSRF, secrets | | Performance | .claude/agents/performance-reviewer.md | Re-renders, bundle, N+1, O(n^2) | | Simplicity | .claude/agents/simplicity-reviewer.md | YAGNI, dead code, over-abstraction | | Next.js/React | .claude/agents/nextjs-react-reviewer.md | Server/Client, hooks, TypeScript, naming |

Launch Pattern:

For each selected reviewer, launch a Task with subagent_type: code-reviewer:

Task: "Run {type} review on {scope}"
Prompt: Read .claude/agents/{agent-file}.md for instructions.
        Review {scope} and produce findings per the agent's output format.
        Include exact file paths and line numbers.

CRITICAL: Launch ALL selected reviewers in a SINGLE message (parallel tool calls). Do NOT run them sequentially.

3. Collect Results

Wait for all agents to complete. Read each agent's output.

4. Compile Report

Create a unified report at:

.agents/reviews/YYYY-MM-DD-{topic}.md

Report Template:

# Code Review: {Topic}

**Date:** YYYY-MM-DD
**Branch:** `{current branch}`
**Scope:** {what was reviewed}
**Reviewers:** {which agents ran}

---

## Executive Summary

{2-3 sentence overview of findings}

| Category | P1 (Must Fix) | P2 (Should Fix) | P3 (Consider) |
|----------|--------------|-----------------|----------------|
| Security | N | N | N |
| Performance | N | N | N |
| Simplicity | N | N | N |
| Next.js/React | N | N | N |
| **Total** | **N** | **N** | **N** |

---

## P1 -- Must Fix

{Findings from all reviewers, merged and deduplicated}

## P2 -- Should Fix

{Findings from all reviewers, merged and deduplicated}

## P3 -- Consider

{Findings from all reviewers, merged and deduplicated}

---

## Positive Findings

{What the codebase does well}

---

## Action Plan

### Immediate (This Session)
- [ ] ...

### Next Sprint
- [ ] ...

### Backlog
- [ ] ...

5. Present Summary

After writing the report, show a compact summary to the user:

REVIEW COMPLETE
----------------------------------------------------------------------

Report: .agents/reviews/YYYY-MM-DD-{topic}.md

Findings:   P1: N | P2: N | P3: N
Reviewers:  security, performance, simplicity, nextjs-react
Scope:      {scope}

Top Issues:
1. {P1 finding summary}
2. {P1 finding summary}
3. {P2 finding summary}

Fix now? (y/n)
----------------------------------------------------------------------

6. Optional: Fix Mode

If the user confirms, start fixing P1 issues:

1. Read each P1 finding
2. Apply the suggested fix
3. Run npx tsc --noEmit to verify
4. Mark fixed in the report

---

Examples

Full Review

/review

Runs all 4 agents on entire codebase.

Security Only

/review security

Runs only the security-reviewer.

Scoped Review

/review frontend/app/api/

Runs all 4 agents scoped to API routes.

---

Integration

• Run /review before /pre-production for comprehensive quality check
• Run /review security after adding new API routes
• Run /review performance after large feature implementations
• Reports accumulate in .agents/reviews/ for historical tracking