Creates new skills or agents from scratch. Invoke proactively when: a repeatable workflow is identified, user says "make a skill", "create an agent", "turn this into a skill", "automate this", "save this workflow", "I keep doing this". If a workflow has been repeated twice, suggest capturing it. Decision: if task needs autonomous multi-step execution → agent; if it's a knowledge/instruction pack → skill.
インストール
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
Creates new skills or agents from scratch. Invoke proactively when: a repeatable workflow is identified, user says "make a skill", "create an agent", "turn this into a skill", "automate this", "save this workflow", "I keep doing this". If a workflow has been repeated twice, suggest capturing it. Decision: if task needs autonomous multi-step execution → agent; if it's a knowledge/instruction pack → skill.
argument-hint
describe the workflow or task to capture
allowed-tools
Read, Write, Edit, Grep, Glob, Bash
Skill: Create
Category: Meta
Role
Creates new skills or agents by analyzing a described workflow and generating the appropriate SKILL.md or agent file with correct frontmatter and trigger phrases.
When to invoke
User says "make a skill", "create an agent", "turn this into a skill"
A workflow has been repeated twice and should be captured
User says "automate this", "save this workflow", "I keep doing this"
A complex multi-step process needs to be standardized
Instructions
Ask clarifying questions: what triggers this workflow? What are the steps? What is the expected output?
Determine type: count agent signals — does it need autonomy, multi-tool use, long-running execution, self-directed decisions? If yes → agent (.claude/agents/.md). If it's a knowledge/instruction pack → skill (.claude/skills//SKILL.md)
Draft the description field first — this is the most critical part. Include all trigger phrases a user might say. Make it specific about auto-activation conditions.
Write the complete file with YAML frontmatter (name, description, argument-hint, allowed-tools)
Include: Role, When to invoke, Instructions, Output format, Example sections
Run a quick sanity check: does the description field contain enough trigger phrases? Would it activate at the right time?
Output format
A complete SKILL.md or agent.md file written to the correct path with no placeholders. Confirmation message with the file path and a brief explanation of the trigger conditions.
Example
/create summarize git commits and post to Slack on every push
