バックエンド実装時に使用。DRY原則遵守。コーディング規約準拠。
using-superpowers
Use when selecting skills for a task or resolving how skill workflows apply
インストール方法を見る含まれるファイル(7)
- SKILL.md2.4 KB
- references/antigravity-tools.md1.5 KB
- references/claude-code-tools.md419 B
- references/codex-tools.md1.1 KB
- references/gemini-tools.md4.5 KB
- references/hermes-tools.md1.9 KB
- references/pi-tools.md1.2 KB
SKILL.md(原文)
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Using Superpowers
Use a skill when the user requests it or its workflow materially helps the task. Read the relevant skill once before applying it, and briefly state its purpose. A matching keyword or a small chance of relevance alone does not require loading a skill. Ordinary questions and read-only exploration can proceed directly.
User instructions and project policy take precedence. Reuse decisions and authorization already present in the conversation; do not restart an approved design or ask again merely because a skill was loaded. Ask only for information or authorization that is still needed. After the design approval, "ask your human partner" in an upstream skill means: decide within the approved design, record a deviation, and report it in the final message; stop only for the exceptions in documents/development/development-policy.md §1.0 "承認後の進め方".
Scale process to the work, not the gate. Every implementation change passes the brainstorming Design Gate (design review, then user approval) before code is written, except the exemption defined there (a typo fix that involves no design choice, or an obviously correct change of a few lines; a fix within an already approved design, such as a quality-check finding, that does not change that design needs no new gate); a bounded change keeps its design short and can then be implemented by the parent with appropriate verification. Use writing-plans for dependencies or a useful handoff. Independent bounded tasks may be delegated when doing so reduces total effort; having subagent tools does not require using them.
The project's quality-check remains the independent final quality gate. Skill workflows do not add per-task reviews or another whole-branch review on top of it. Keep the project's review budget, test oracles, and evidence requirements.
For tool syntax consult only the reference relevant to your runtime, and trust the actual available tools and instructions over examples:
- Claude Code: references/claude-code-tools.md
- Codex: references/codex-tools.md
- Pi: references/pi-tools.md
- Antigravity: references/antigravity-tools.md
- Hermes: references/hermes-tools.md
レビュー
まだレビューはありません。使ってみた感想をお寄せください。
同じリポジトリのスキル
概要と使いどころ
Use before implementing any feature, behavior change, or refactor - settles requirements and design, then gets the design independently reviewed and approved by the user before code is written
日本語の概要は準備中です。原文の説明を表示しています。
作業開始時に使用。mainブランチでの作業禁止。Issue先行作成必須。
UI実装後の検証時に使用。agent-browser CLIでブラウザ上の動作を手動検証する。「UIを確認」「画面テスト」と言われたら使用(プロジェクトの E2E スイート実行は quality-check Step 5(推奨度・範囲で自動実施または確認)/ server-startup が担当)。
DBマイグレーション作成時に使用。バージョン番号競合防止。mainブランチ確認必須。
Use when independent tasks benefit from parallel work without shared state or sequential dependencies
日本語の概要は準備中です。原文の説明を表示しています。