---
name: genius-code-review
description: >-
  Multi-agent PR code review. Dispatches 3 parallel reviewers: bugs finder, security
  auditor, and code quality reviewer. Produces a single consolidated review report.
  Inspired by Anthropic's Code Review system (84% bug detection on large PRs).
  Use when user says "review this PR", "code review", "review pull request",
  "audit the code changes", "multi-agent review".
  Do NOT use for general code quality suggestions — use genius-reviewer for that.
context: fork
agent: genius-code-review
user-invocable: true
allowed-tools:
  - Read(*)
  - Write(*)
  - Edit(*)
  - Glob(*)
  - Grep(*)
  - Bash(git diff*)
  - Bash(git status*)
  - Bash(git log*)
  - Bash(git show*)
hooks:
  PostToolUse:
    - type: command
      command: "bash -c 'echo \"[$(date +%H:%M:%S)] CODE-REVIEW: $TOOL_NAME\" >> .genius/review.log 2>/dev/null || true'"
  Stop:
    - type: command
      command: "bash -c 'echo \"CODE-REVIEW COMPLETE: $(date)\" >> .genius/review.log 2>/dev/null || true'"
      once: true
---

# Genius Code Review v17 — Multi-Agent PR Reviewer

**Three reviewers. One report. Zero bugs missed.**

Inspired by Anthropic's multi-agent code review research showing **84% bug detection** on large PRs — versus ~20% for single-pass review.

---

## Mode Compatibility

| Mode | Behavior |
|------|----------|
| **CLI** | Full parallel review; diff extracted via `git diff`; report written to `.genius/reviews/` |
| **IDE** (VS Code/Cursor) | Same as CLI; open report in editor panel after completion |
| **Omni** | Each reviewer agent uses the best available model for its specialty |
| **Dual** | Claude leads bugs + security; Codex handles code quality patterns review |

---

## Architecture: 3 Parallel Reviewers

```
                  ┌─────────────────────┐
                  │   genius-code-review │
                  │   (orchestrator)     │
                  └──────────┬──────────┘
                             │ dispatch in parallel
           ┌─────────────────┼─────────────────┐
           ▼                 ▼                 ▼
    ┌─────────────┐  ┌──────────────┐  ┌─────────────────┐
    │ bugs-finder │  │  security-   │  │ quality-reviewer │
    │             │  │  scanner     │  │                  │
    └──────┬──────┘  └──────┬───────┘  └────────┬────────┘
           └────────────────┼──────────────────┘
                            │ merge findings
                    ┌───────▼──────────┐
                    │ Consolidated     │
                    │ Review Report    │
                    └──────────────────┘
```

---

## Step 1: Get the Diff

```bash
# PR diff (against main)
git diff main...HEAD > /tmp/pr.diff

# Or specific commit
git show <commit-sha> > /tmp/pr.diff

# Or staged changes
git diff --staged > /tmp/pr.diff

# Stats overview
git diff main...HEAD --stat
```

Read the diff to understand scope: files changed, lines added/removed, nature of changes.

---

## Step 2: Dispatch 3 Reviewers in Parallel

In Agent Teams mode, spawn all 3 simultaneously:

**Reviewer 1 — Bug Finder**: Logic errors, edge cases, null handling, race conditions, async misuse, boundary conditions, type mismatches, error handling. Output: `BUGS.md` (severity + file:line + explanation + fix).

**Reviewer 2 — Security Scanner**: SQL injection, XSS, hardcoded secrets, missing auth, IDOR, input validation, path traversal, CORS, rate limiting, weak crypto, webhook verification. Output: `SECURITY.md` (OWASP category + severity + file:line + remediation).

**Reviewer 3 — Code Quality**: DRY violations, SRP, test coverage, naming, magic numbers, complexity, N+1 queries, dead code, missing docs, breaking changes. Output: `QUALITY.md` (category + impact + file:line + suggestion).

---

## Step 3: Consolidate into Final Report

Merge all three reviewer outputs into a ranked report with these sections:
- Header: PR name, date, files/lines changed
- **🔴 CRITICAL** — must fix before merge (each: `[C1]` ID, file:line, reviewer, category, issue, fix)
- **🟠 HIGH** — significant bugs/security (same format with `[H1]` IDs)
- **🟡 MEDIUM** — quality/tech debt issues
- **🟢 LOW** — minor improvements, style, docs
- **Summary table**: severity × count × must-fix-before-merge
- **Verdict**: ✅ Approved / ⚠️ Approved with changes / ❌ Changes required

---

## Step 4: Save and Deliver Report

```bash
# Save report
mkdir -p .genius/reviews
REPORT_PATH=".genius/reviews/review-$(date +%Y%m%d-%H%M%S).md"
# Write consolidated report to $REPORT_PATH

echo "Review saved: $REPORT_PATH"
cat $REPORT_PATH
```

---

## CLI / Non-Multi-Agent Mode

Without Agent Teams, run three sequential passes: bugs, security, and maintainability. Consolidate them into the standard report format.

---

## Output

Mark `.genius/outputs/state.json` complete for `genius-code-review` with a fresh timestamp.

---

## Handoff

- → **genius-security**: Deep security audit if CRITICAL issues found
- → **genius-qa-micro**: Generate tests to cover bug scenarios found
- → **genius-reviewer**: Ongoing code quality coaching (non-PR specific)

---

## Playground Update

Refresh the existing dashboard tab with real review data and tell the user to open `.genius/DASHBOARD.html`.


---

## Output Quality Requirements

Every issue must include file:line, severity, category, description, and a concrete fix. The summary must include counts by severity, an overall assessment, and a go/no-go call. If no issues are found, say so and name the observed strengths.

## Definition of Done

- [ ] Review covers changed files and user-stated scope
- [ ] Findings include file references, severity, and actionable fixes
- [ ] Security, regression, and test gaps evaluated where relevant
- [ ] Overall go/no-go recommendation is explicit
- [ ] Dashboard tab or status output updated if the workflow expects it