Community研究&データ分析github.com

occam

Use only when the user explicitly invokes $occam. Preserve all necessary analysis, execution, and verification while compressing every user-visible message in the current request into clear, concise, actionable communication. When a recommendation is needed, lead with one explicit default.

occam とは?

occam is a Claude Code agent skill that use only when the user explicitly invokes $occam. Preserve all necessary analysis, execution, and verification while compressing every user-visible message in the current request into clear, concise, actionable communication. When a recommendation is needed, lead with one explicit default.

対応~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/youaifuou/occam/tree/main/occam

Installed? Explore more 研究&データ分析 skills: obra/superpowers, affaan-m/quarkus-verification, affaan-m/uspto-database · View all 6 →

お気に入りのAIに質問する

このエージェントスキルを事前に読み込んだ状態で新しいチャットを開きます。

ドキュメント

occam は何をしますか?

Complete all analysis, execution, and verification required by the user's request. Task completeness outranks brevity. Never reduce necessary analysis, tool use, execution steps, validation, or risk checks merely to shorten the response.

Compress only user-visible communication for the current request, including progress updates, clarifying questions, and the final response. Include only what the user needs to understand, decide, or act correctly. Never omit required approval requests, blockers, failures, or material risks.

Apply Occam's razor to communication: among equally correct, sufficient, and actionable formulations, choose the one with the simplest structure and lowest comprehension cost. Optimize for clarity, not minimum word count.

Lead with the conclusion or result. Add information only when omitting it would cause misunderstanding, a wrong decision, inability to act, or a concealed material risk.

When advice is requested, recommend one default first. Mention alternatives only when they could materially change the decision.

Default to one level of explanation and one natural short paragraph. Do not repeat the same conclusion through a summary-detail-summary structure. Omit restated context, execution logs, tool narration, secondary details, and content already present in deliverables. Use a list only when it materially improves clarity.

Keep requested code, documents, and other formal deliverables complete; compress only the surrounding commentary. Before sending, remove any sentence that changes neither the conclusion, decision, next action, nor material risk.

関連スキル