メニュー
Review code with grug-brain anti-complexity philosophy. Use when reviewing PRs, diffs, or code changes to identify unnecessary complexity.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Create logically grouped, atomic git commits with well-formatted commit messages following best practices. Use this skill when you need to commit changes to a git repository with proper message formatting and atomic grouping. Before creating commits, reflect on the work just completed and surface any open questions or unresolved decisions (judgment calls made without checking, deferred edge cases, defaults picked when multiple options were viable, trade-offs that could reasonably go the other way) via AskUserQuestion so they get discussed with the user rather than silently committed.
Implement an epic's beads in a worktree with logical commits, multi-agent review, fix pipeline, and PR description generation. Use when you want to implement this epic, burn down the beads, execute the plan, or start working on an epic.
Comprehensive branch review using parallel background agents with Codex validation. Each reviewer runs as an independent background task - results are collected only after all agents complete, eliminating premature synthesis. Use for thorough pre-PR review of any branch size.
Implement fixes for code review findings using a parallel agent team. Accepts a review report (from /team-branch-review, /team-commit-review, or pasted), interviews the user on which findings to fix (or runs autonomously with --auto), then spawns agents to implement fixes in parallel with Codex validation.
Comprehensive branch review using a parallel agent team with Codex validation. Use when reviewing all commits on a branch before creating a PR, especially for large or complex branches that benefit from multi-perspective review. Trigger on "review my branch", "check before I PR", "full code review", "pre-PR review", "team review", or "review before merge".
Comprehensive commit-scoped review using a parallel agent team with Codex validation. Use when reviewing specific commits rather than an entire branch - ideal for auditing a single commit, the last N commits, or a specific SHA before or after merging. Trigger on "review this commit", "review last 3 commits", "review HEAD", "commit review", or "check this commit".
| name | grug-review |
| description | Review code with grug-brain anti-complexity philosophy. Use when reviewing PRs, diffs, or code changes to identify unnecessary complexity. |
| argument-hint | [--epic <EPIC_ID>] |
Review code changes through the grug-brain lens. Complexity is the enemy.
Before running the checklist, follow ~/.claude/skills/shared/planning-context.md to resolve an epic (via --epic <EPIC_ID> flag, branch name parse, or fallback) and produce a planning-context block. Surface the block at the top of your output — reviewers and the user see the same intent signal the code authors worked against.
If the loader falls back to "No linked planning context available — reviewing against general code-quality heuristics only.", print that line and proceed with the checklist normally.
When planning context exists, treat explicit design decisions as intentional: complexity the plan chose is not automatically "slop." Flag it only if the complexity exceeds what the decision justifies.
For each change, ask:
Structure your review as:
### Complexity Concerns
**Premature abstraction in `DataHandler`**: This abstract base class has only one implementation. Grug say: delete the ABC, keep the concrete class. Abstract when second implementation actually needed.
**Over-engineered error handling**: The retry decorator with exponential backoff handles errors that cannot happen in this context. Grug say: remove decorator, add simple try/except only if error is possible.
### Suggested Simplifications
The `validate_and_transform_and_persist` method does three things. But grug note: these three things always happen together, in this order, nowhere else. Splitting into three methods adds indirection without benefit. Keep as one method, maybe rename to `save_validated`.
### Acceptable Complexity
The caching layer adds complexity but profiler shows 40% latency reduction. Complexity justified by measurement. Grug approve.