スキル multi-agent-brainstorming
📦

multi-agent-brainstorming

安全

Multi-Agent Design Review

Single-agent designs often miss hidden assumptions and failure modes. This skill runs a structured, sequential review with five specialized agents to validate designs before implementation.

対応: Claude Codex Code(CC)
🥉 75 ブロンズ
1

スキルZIPをダウンロード

2

Claudeでアップロード

設定 → 機能 → スキル → スキルをアップロードへ移動

3

オンにして利用開始

テストする

「multi-agent-brainstorming」を使用しています。 Use multi-agent-brainstorming to review a new API design for a payment processing system

期待される結果:

  • Design Review Complete
  • Decision Log: 5 decisions recorded
  • Objections Addressed: 8 reviewer objections resolved
  • Final Disposition: APPROVED with 2 required revisions
  • Key Findings: Security constraints added for PII handling, scalability requirements clarified

「multi-agent-brainstorming」を使用しています。 Review this database schema design using multi-agent-brainstorming

期待される結果:

  • Design Review Complete
  • Decision Log: 4 decisions recorded
  • Objections Addressed: 6 reviewer objections resolved
  • Final Disposition: REVISE
  • Key Findings: Performance concerns on query patterns, denormalization needed for read-heavy flows

セキュリティ監査

安全
v1 • 2/25/2026

All static findings are false positives. The skill is a legitimate multi-agent design review framework. Detected patterns are markdown formatting (backticks for skill references) and design review terminology, not actual security threats. No malicious code, commands, or cryptographic algorithms present.

1
スキャンされたファイル
262
解析された行数
2
検出結果
1
総監査数
中リスクの問題 (1)
SKILL.md:52SKILL.md:159
Markdown Backtick False Positive
Static scanner detected 'Ruby/shell backtick execution' at SKILL.md:52 and 159. These are markdown backticks referencing the 'brainstorming' skill, not shell commands.
低リスクの問題 (1)
SKILL.md:3-4SKILL.md:210SKILL.md:255
Terminology Misinterpretation
Scanner detected 'weak cryptographic algorithm' and 'system reconnaissance' patterns. These are false positives from design review terminology in the document text.
監査者: claude

品質スコア

38
アーキテクチャ
100
保守性
87
コンテンツ
50
コミュニティ
98
セキュリティ
100
仕様準拠

作れるもの

Critical System Design Validation

Validate architectural decisions for production systems where failure has significant consequences.

Feature Design Peer Review

Review new feature proposals before development begins to catch edge cases and usability issues.

Technical Decision Documentation

Create a structured Decision Log that documents all design choices and their rationale.

これらのプロンプトを試す

Basic Design Review 
Use the multi-agent-brainstorming skill to review this design: [describe your design idea]. Run the full review process and produce a complete Decision Log.
Security-Focused Review 
Use the multi-agent-brainstorming skill to review [describe design]. Focus the Constraint Guardian agent specifically on security and privacy aspects.
User Experience Review 
Use the multi-agent-brainstorming skill to review [describe design]. Emphasize the User Advocate agent to evaluate user experience and usability.
Quick Feasibility Check 
Use the multi-agent-brainstorming skill for a rapid review of [describe design]. Focus on identifying showstoppers and major risks only.

ベストプラクティス

  • Provide clear, specific design requirements to the Primary Designer before starting review
  • Ensure all five agent roles are invoked in sequence for comprehensive coverage
  • Complete the Decision Log for every design review as mandatory documentation

回避

  • Skipping the Skeptic agent - challenging assumptions is critical for finding hidden flaws
  • Allowing the Primary Designer to self-approve without Integrator review
  • Exiting the review process before all exit criteria are met

よくある質問

What is the difference between multi-agent-brainstorming and regular brainstorming?
Regular brainstorming is parallel and creative. Multi-agent-brainstorming is sequential and critical. It enforces structured review with specific agent roles to find flaws, not generate ideas.
Can I skip some agent roles to save time?
No. Each agent role serves a specific purpose. Skipping roles violates the skill's exit criteria and reduces review quality.
What happens if an objection cannot be resolved?
The Integrator must explicitly reject the objection with rationale, documenting it in the Decision Log. The design cannot proceed without this explicit rejection.
How do I know when the review is complete?
The review is complete when: Understanding Lock is done, all reviewers invoked, all objections resolved or rejected, Decision Log complete, and Arbiter declares acceptable.
Can this skill work with any Claude or Codex setup?
Yes, the skill works with any Claude, Codex, or Claude Code environment that can manage multi-turn conversations with defined agent roles.
What if I need to reopen a completed design decision?
The Integrator must provide explicit cause for reopening. Without cause, locked decisions remain final per the skill's non-negotiable rules.

開発者の詳細

作成者

sickn33

ライセンス

MIT

リポジトリ

https://github.com/sickn33/antigravity-awesome-skills/tree/main/skills/multi-agent-brainstorming

参照

main

ファイル構成

📄 SKILL.md