メニュー
Use for ad-hoc code review: after a major feature, before merging to main, or when stuck. Executing a written plan instead? Use executing-plans, which owns the per-task review cycle.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
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。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/carlos-algms/dotfiles --skill requesting-code-reviewコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
Use when a `discussion_rNNN` URL anchor appears, when replying to a specific PR review comment from a bot or AI reviewer (CodeRabbit, cubic, Claude, Copilot) or a human reviewer, reading all reviews on a PR with their replies as a tree, or resolving review threads from the CLI. Also covers where AI reviewers put out-of-diff findings and the mandatory `@handle` tag when replying. Triggers on "reply to coderabbit", "reply to cubic", "reply to the bot", "answer the bot review", "address bot reviews", "answer AI review comments", "answer this review comment", "read all reviews on PR", "show review threads", "resolve this thread".
Execute and orchestrate multiple AI agent CLIs (agy, Codex, Cursor, Claude Code, Pi) for getting alternative opinions, comparing approaches, or leveraging different AI models' strengths. Use when the user explicitly mentions agent names like "ask agy", "ask cursor", "ask codex", "ask pi", "use composer", "use pi", "try grok", or when needing diverse perspectives on complex problems, alternative implementation approaches, or multi-agent collaboration.
Compose and create GitHub pull requests with `gh` using the target repository's title, template, base-branch, and draft conventions. Load on explicit user intent to open, create, draft, or compose a pull request.
SOC 職業分類に基づく
SKILL.md を表示中
| name | requesting-code-review |
| description | Use for ad-hoc code review: after a major feature, before merging to main, or when stuck. Executing a written plan instead? Use executing-plans, which owns the per-task review cycle. |
Dispatch a fresh reviewer with pointers, not session history.
This skill covers AD-HOC review: work not driven by a plan file.
Executing a plan? Stop and use executing-plans.
Use ad-hoc review before merge, after a major unplanned feature, or when stuck.
git rev-parse HEAD
git status --porcelain
BASE_REF - commit before the complete review scope. Resolve it from the
target branch or supplied starting commit; use current HEAD only for purely
uncommitted workCHANGED_FILES - deduplicated, newline-delimited exact-path union of
committed changes since BASE_REF and current staged, unstaged, deleted, and
untracked changes. Use rename destinations. These are entry points, not a
boundaryMUST read instructions at <skill_dir>/code-reviewer.md FIRST.
Do not act until you have read it. Then apply:
plan_or_requirements = <requirements or plan reference>
base_ref = <commit before the work>
baseline_snapshot = none
changed_files = <newline-delimited exact paths>
solved_defects = none
Resolve <skill_dir> to the directory containing this SKILL.md.
Ad-hoc review has no baseline snapshot and no task scope, so baseline_snapshot
is always none and task_id and scope_mode are omitted. The reviewer
reviews the full changed_files delta. Plan execution supplies those values
instead.
PASS: finishReject incorrect findings only with code or test counter-evidence. Use one
clarification round for ambiguity; escalate unresolved disputes. Keep reviewer
responses out of the parent context except for PASS or the short finding list.
Allow 3 total attempts for empty, errored, or malformed output. Limit findings
to 3 fix/re-review rounds; escalate before a fourth.