skill-authoring-template
新しいSkillを作るときの必須セクション(入力/出力/禁止/検証/根拠)テンプレ。skill-builder が新規Skillを生成するときに参照する。
When & Why to Use This Skill
This Claude skill provides a standardized framework and mandatory template for authoring new agent skills. By defining essential sections such as Inputs, Outputs, Forbidden actions, and Verification steps, it ensures that all developed skills are consistent, reliable, and optimized for agentic workflows. It serves as a foundational guide for developers and automated skill-builders to maintain high-quality documentation and naming conventions within an AI ecosystem.
Use Cases
- Standardizing Skill Development: Use the template to ensure all new Claude skills follow a uniform structure, making them easier to maintain and audit.
- Enhancing Agent Reliability: Define explicit 'Forbidden' and 'Verification' sections to minimize hallucinations and ensure the AI follows strict operational boundaries.
- Automated Skill Generation: Provide this template as a reference for 'skill-builder' agents to programmatically generate valid and well-structured SKILL.md files.
- Optimizing Metadata for Discovery: Implement consistent YAML frontmatter and naming patterns to improve the discoverability and integration of skills within complex agent orchestrators.
| name | skill-authoring-template |
|---|---|
| description | 新しいSkillを作るときの必須セクション(入力/出力/禁止/検証/根拠)テンプレ。skill-builder が新規Skillを生成するときに参照する。 |
Skill Authoring Template
YAML frontmatter (MUST)
SKILL.md は必ず YAML フロントマターから始める。
最低限:
- name
- description(「何をする」+「いつ使う」+「キーワード」)
禁止: description: | のようなマルチライン構文は使用しないこと。
Required sections (MUST)
- Inputs
- Outputs
- Forbidden
- Verification
- Evidence (which files/dirs informed this)
Naming (MUST)
- 小文字/数字/ハイフンのみ、64文字以内
- 予約語や紛らわしい語を避ける(例: anthropic / claude を含めない)
- 推奨: 一貫した命名パターン(公式は gerund: 動名詞形 “verb-ing” を推奨)
例(gerund):
testing-codewriting-documentationmanaging-migrations
Guidance
- 手順は「誰がやっても同じ」粒度で具体的に
- 検証コマンドは repo に実在するものだけを書く