| name | groom |
| description | Multi-perspective code review using parallel subagents. Use when asked to 'groom', 'review my changes', 'clean up code', 'check code quality', or after completing a major feature. |
| argument-hint | [file or directory to focus on] |
/groom — Multi-Perspective Code Review
Get the git diff of recent changes, classify the change, then launch parallel review subagents. Synthesize their feedback into a prioritized report.
Reviewers
Launch all applicable reviewers simultaneously as subagents.
1. CLAUDE.md Compliance (always run)
The subagent should read all CLAUDE.md files (user-level, project-level, repo root) and check every change against documented guidelines. Flag violations with specific file:line references.
If you are Codex, then check for AGENTS.md compliance instead. (Often AGENTS.md is aliased to CLAUDE.md, but if both exist, check AGENTS.md for Codex and CLAUDE.md for Claude.)
2. Complexity Reducer (always run)
Apply Software Design Principles to identify unnecessary complexity:
- Deep modules over shallow — Flag shallow wrappers, interfaces as complex as implementations
- Pull complexity downward — Callers shouldn't handle complexity that belongs inside a module
- Define errors out of existence — Error handling eliminable through better API design
- Different layer, different abstraction — Flag pass-through methods adding no value
- Information hiding — Implementation details leaking through interfaces
- General-purpose modules are deeper — Overly specific solutions with simpler general alternatives
- Avoid excessive decomposition — Many small functions/classes that could be fewer, deeper abstractions
Also fix other basic code quality issues: naming, formatting, comments, dead code, duplication.
For each issue: state the problem, which principle it violates, and a concrete fix.
3. Test Coverage (conditional: if touching testable code or tests)
Check if new code paths are tested. Are tests testing behavior (not implementation)? Obvious edge cases missing? Brittle coupling to implementation details?
4. Performance, Correctness & Security (conditional: if involving data processing, complex logic, I/O, auth, or user input)
Performance (O(n²) where O(n) possible, unnecessary copies). Correctness (off-by-one, race conditions, unhandled edge cases). Security (injection, hardcoded secrets, unsafe deserialization).
Output Format
Combine all findings into a deduplicated, prioritized report:
## 🐱 Groom Report
### Critical (must fix)
- [file:line] Issue — Reviewer: X
### Recommended (should fix)
- [file:line] Issue — Reviewer: X
### Suggestions (nice to have)
- [file:line] Issue — Reviewer: X
### ✅ Looks Good
- Aspects that passed with no issues