Skip to main content

write-a-skill

Guides creation of new agent skills with correct structure, wiring, and tests. Use when creating a new skill for the .github/skills/ framework.

ソース情報

リポジトリ
wryenmeek/knowledgebase
ソースの最終更新活動
2026年4月25日 16:37
検出された SKILL.md の言語
英語
スター
1
フォーク
0

インストール方法

デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。

ソースファイルを確認

インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。

SKILL.md を表示中

SKILL.md
ソースの指示 · 読み取り専用プレビュー
name
write-a-skill
description
Guides creation of new agent skills with correct structure, wiring, and tests. Use when creating a new skill for the .github/skills/ framework.
# Write a Skill ## Overview <!-- Classification: Doc-only workflow --> Step-by-step procedure for creating new skills that integrate correctly with the framework's discovery, routing, and testing infrastructure. ## When to Use - Creating a new skill for `.github/skills/` - Porting an external skill pattern into this framework - Unsure about required skill structure or wiring steps ## Procedure ### 1. Check for overlap Search `using-agent-skills/SKILL.md` discovery tree. If an existing skill covers the use case, extend it instead of creating a new one. ### 2. Choose routing category - **Operator-direct**: Developer invokes directly during development workflow (e.g., `zoom-out`, `grill-me`). No persona owner needed. - **Persona-routed**: Agent persona invokes during governance pipeline (e.g., `enforce-npov`). Must be listed in the persona's Required skills. ### 3. Create skill directory and SKILL.md ``` .github/skills/<skill-name>/SKILL.md ``` Required frontmatter: - `name`: must match directory name exactly - `description`: must contain "Use when..." trigger phrase Required sections: - `## Overview` — one-paragraph summary - `## When to Use` — bullet list of trigger conditions - `## Procedure` or `## Rules` — step-by-step workflow or rule set - `## Verification` — how to confirm the skill was applied correctly Size target: ≤ 100 lines. ### 4. Add logic (if write-capable) If the skill performs repository writes: 1. Create `logic/` subdirectory with Python module 2. Follow `_optional_surface_common.py` pattern (`SurfaceResult`, `JsonArgumentParser`, `run_surface_cli`) 3. Declare in AGENTS.md write-surface matrix 4. Add lock requirements and hard-fail conditions to SKILL.md ### 5. Wire into framework 1. Add to `using-agent-skills/SKILL.md` discovery tree 2. Add to `.github/copilot-instructions.md` lifecycle mapping and intent mapping 3. If persona-routed: add to the relevant agent persona's Required skills ### 6. Update tests 1. `tests/kb/test_framework_skills.py` — skill name appears in framework validation 2. If write-capable: `tests/kb/test_framework_write_surface_matrix.py` — matrix row exists ## Verification 1. `name` in frontmatter matches directory name. 2. `description` contains "Use when". 3. File is ≤ 100 lines. 4. Skill appears in `using-agent-skills` discovery tree. 5. `python3 -m pytest tests/kb/test_framework_skills.py -v` passes.
GitHubで見る