{
  "markdown": "# devils-advocate\n\n> A **devil's advocate agent** for Claude Code. Place it at the end of any plan / design / multi-agent review pipeline to break groupthink — with the strict rule: *only fatal points, and every point must come with a fix*.\n\n[English](#english) | [日本語](#日本語)\n\n---\n\n## English\n\n### What does it do?\n\nRun `/devils-advocate` and the agent reviews the most recent plan, design, or review output and returns:\n\n- **Fatal points only** (no nitpicks)\n- **Every point paired with a concrete fix or alternative** (no pure negativity)\n- **Numbers, facts, and specifics** (no vibes)\n- A final verdict: *Agree / Conditional Agree / Object*\n- A forward-looking closing line: *\"Fix this and it becomes the real thing\"*\n\n### Why this exists\n\nParallel agents and review panels share the same priors, so they share the same blind spots:\n\n- The plan everyone agreed on quietly fails in production\n- The design no one objected to dies on the first user touchpoint\n- The multi-agent synthesis report is drenched in optimism bias\n\nThis plugin is the *last gate* — a single command that injects an opposing perspective right before you ship the decision.\n\n### Design principles\n\n#### Do\n- Surface only **fatal** problems\n- Pair every problem with a **fix or alternative**\n- Strike with **numbers, facts, and concrete examples**\n\n#### Don't\n- Pure negativity (\"this won't work\")\n- Pedantic nitpicks, style critique, preference enforcement\n- Problems without proposed fixes\n\n### Review axes (general checklist)\n\n| # | Axis |\n|---|---|\n| A | Plan / design itself |\n| B | Validity of expected outcomes |\n| C | User / recipient blind spots |\n| D | Execution and ops reality |\n| E | Technical implementation pitfalls |\n| F | Legal / ethics / policy |\n| G | Risk management |\n\n### Installation\n\n#### From the official marketplace (in review)\n\n```\n/plugin install devils-advocate\n```\n\n#### Directly from GitHub\n\n```bash\ngit clone https://github.com/keitashimoda24-maker/devils-advocate.git \\\n  ~/.claude/plugins/devils-advocate\n```\n\nThen run `/reload-plugins` in Claude Code.\n\n### Usage\n\nOnce a plan, design, or integrated report exists in the conversation:\n\n```\n/devils-advocate\n```\n\nOutput format:\n\n- Up to 3 fatal points, each with a fix\n- Optimism bias warnings\n- Blind-spot angles everyone missed\n- Final verdict\n- Forward-looking single closing line\n\n### Recommended use cases\n\n- Final check on multi-agent integrated reports\n- Second opinion before major decisions\n- An additional pass on top of standard code/design reviews\n- Self-review of your own plans and proposals\n\n---\n\n## 日本語\n\n### 何ができる？\n\n`/devils-advocate` を実行すると、直前の計画・設計・レビュー結果に対して以下を返します:\n\n- **致命傷レベルの指摘のみ**（重箱の隅はNG）\n- 各指摘に**必ず改善案 or 代替案がセット**で付くので、単なるダメ出しで終わらない\n- **数字・事実・具体例で殴る**（感情論禁止）\n- 最終判定（同意 / 条件付き同意 / 異議あり）\n- 前向きな最終1行（「ここを直せば本物になる」）\n\n### なぜ作ったか\n\n並列稼働する複数の専門家・エージェントは、**同じ前提から発想するため集団思考バイアスに陥りがち**です。\n\n- 全員が「いける」と言った企画が大コケする\n- レビュー会で誰も反対しなかった設計が本番でハマる\n- マルチエージェントの統合レポートが楽観バイアスまみれになる\n\n→ このプラグインは「最後の砦としての否定役」を1コマンドで起動します。\n\n### 設計思想\n\n#### Do\n- **致命傷レベルの指摘のみ**（重箱の隅はNG）\n- 各指摘に**必ず改善案を添える**\n- 数字・事実・具体例で殴る（感情論禁止）\n\n#### Don't\n- 単なる否定で終わる「ダメ出し」\n- 揚げ足取り・文体批判・好みの押しつけ\n- 改善案なしの「これは無理」\n\n### レビュー観点（汎用チェックリスト）\n\nA. 企画/設計そのものへの疑い\nB. 想定成果の妥当性\nC. ユーザー/受け手視点の罠\nD. 実行・運用の現実\nE. 技術・実装の落とし穴\nF. 法務・倫理・規約\nG. リスク管理\n\n### インストール\n\n#### 公式マーケットから（レビュー待ち）\n\n```\n/plugin install devils-advocate\n```\n\n#### GitHub から直接\n\n```bash\ngit clone https://github.com/keitashimoda24-maker/devils-advocate.git \\\n  ~/.claude/plugins/devils-advocate\n```\n\nその後 Claude Code で `/reload-plugins` を実行。\n\n### 使い方\n\nレビュー対象（計画書・設計書・統合レポート等）が会話に存在する状態で:\n\n```\n/devils-advocate\n```\n\nエージェントが起動し、以下フォーマットで出力します:\n\n- 致命傷指摘（最大3つ、改善案セット）\n- 楽観バイアスへの警告\n- 全員が見落とした角度\n- 最終判定（同意 / 条件付き同意 / 異議あり）\n- 前向きな最終1行\n\n### 推奨ユースケース\n\n- マルチエージェント統合レポートの最終チェック\n- 重要な意思決定前のセカンドオピニオン\n- コードレビュー・設計レビュー結果への上乗せレビュー\n- 自分が書いた企画書・提案書のセルフチェック\n\n---\n\n## Background\n\nThis pattern came out of running multi-agent pipelines for real operational decisions (shift scheduling, payroll distribution, expense audits) across 10 convenience stores. Every time I added another \"specialist\" agent, the integrated report got *more* confident — not more correct. The fix wasn't a smarter specialist; it was a structurally adversarial final gate.\n\n## Author\n\n[@keitashimoda24-maker](https://github.com/keitashimoda24-maker)\n\nNon-engineer area manager running 10 convenience stores in Japan, building daily automation with Claude Code (30+ systems and counting).\n\n## Related Projects\n\n- [claude-code-smart-hooks](https://github.com/keitashimoda24-maker/claude-code-smart-hooks) — Conditional ultrathink injection for consistent reasoning depth\n- [slash-commands-jp](https://github.com/keitashimoda24-maker/slash-commands-jp) — Japanese slash command reference plugin\n- [mac-pixel-bridge](https://github.com/keitashimoda24-maker/mac-pixel-bridge) — Resilient ADB clipboard / screenshot bridge\n\n## License\n\nMIT — see [LICENSE](LICENSE).\n",
  "bytes": 4887,
  "sha": "21ff5e8915389e507f555253f8d00449dfcb9b99ea6b8d51f9bf3c65b8514fa9",
  "repo_slug": "keitashimoda24-maker/devils-advocate",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_keitashimoda24_maker_devils_advocate_dev_28245b1a/readme"
}