review

name	review
description	Multi-agent parallel code review against the base branch
user-invocable	true
allowed-tools	["Read","Glob","Grep","Agent","Bash(git diff:*)","Bash(git log:*)","Bash(git status)","Bash(gh pr view:*)"]

/review — Multi-Agent Code Review

Review the current branch's changes against the base branch using parallel review agents.

Step 0: Prepare

1. Identify base branch (main or master) and current branch
2. git log — commits on this branch
3. git diff {base}...HEAD — full changeset
4. gh pr view --json body — PR description (if available)
5. Run each command individually — do NOT chain with &&

Step 1: Build Change Context

1. Understand the intent from commit messages and PR description
2. Categorize changed files: core logic, UI, infrastructure, tests
3. Identify risk areas: complex changes, new integrations, security-sensitive code

Step 2: Multi-Agent Parallel Review

Launch two review agents in parallel (model: sonnet):

Agent A: Bug & Logic + Security

Review the changed files for bugs, logic errors, and security issues.
Read REVIEW.md (if exists) or .claude/rules/ for review criteria.

Focus areas:
- Null safety violations
- Concurrency issues
- Error handling gaps
- Security: hardcoded secrets, input sanitization, data exposure
- Platform compatibility
- Performance: memory leaks, main thread blocking, redundant calls

For each finding: [file:line] severity — description
Severity: Critical / Warning / Suggestion / Nit

Agent B: Architecture & Quality

Review the changed files for architecture and code quality.
Read REVIEW.md (if exists) or .claude/rules/ for review criteria.

Focus areas:
- Architecture: SRP, layer boundaries, dependency direction
- Design patterns: consistency with existing codebase
- Testing: changed code has corresponding test updates
- Naming and readability
- Unnecessary complexity or over-engineering

For each finding: [file:line] severity — description
Severity: Critical / Warning / Suggestion / Nit

Project-Specific Reviewers (if available)

Check .claude/agents/ for project-specific reviewer agents (e.g., kmp-reviewer, ui-reviewer). If found, launch them in parallel with Agents A and B.

Step 3: Merge Findings

1. Collect findings from all agents
2. Deduplicate: remove findings reported by multiple agents
3. Assign final severity:
   • Critical: crash, data loss, security vulnerability, incorrect behavior
   • Warning: potential bug, performance issue, architecture violation
   • Suggestion: improvement opportunity, non-blocking
   • Nit: style/preference, optional

Step 4: Present Report

## Review Summary

**Branch:** {current} → {base}
**Files changed:** N
**Risk level:** Low / Medium / High
**Reviewed by:** Agent A (Bug/Security) + Agent B (Arch/Quality) + {project-specific}

### Critical (must fix)
- [file:line] description

### Warning (should fix)
- [file:line] description

### Suggestion (nice to have)
- [file:line] description

### Nit (optional)
- [file:line] description