Create Skill

Ask one question at a time using ask_user:

1. Invocation: "Who should invoke this skill?"

   • "Both user and Claude (default)"
   • "User only (disable-model-invocation: true)"
   • "Claude only (user-invocable: false)"

2. Execution context: "Where should this skill run?"

   • "Inline in main conversation (default)"
   • "In a forked subagent (context: fork)"

3. If context: fork was chosen, ask agent type:

   • "general-purpose (default)"
   • "Explore (read-only, fast)"
   • "Plan (read-only research)"
   • "Custom agent name"

4. Tool restrictions: "Restrict tools for this skill?"

   • "No restrictions (default)"
   • "Let me specify allowed tools" If user wants to specify, ask for a comma-separated list.

5. Arguments: "Does this skill accept arguments?"

   • "No"
   • "Yes — let me provide an argument hint" If yes, ask for the hint (e.g. [filename], [issue-number]).

Step 5: Generate the Skill

Use the template in templates/skill-template.md to generate the skill directory. Key rules:

• Create <scope-path>/<skill-name>/SKILL.md with YAML frontmatter + markdown body
• Only include frontmatter fields that differ from defaults
• description is always included
• If the skill type is "Task", include numbered step placeholders
• If the skill type is "Reference", include bullet-point convention placeholders

For supporting files:

• Ask: "Add supporting files?" with choices: "No", "Yes — reference.md", "Yes — reference.md + scripts/"
• If yes, create stub files with TODO comments

Step 6: Summary

Print:

✅ Skill "<name>" created at <path>/<name>/SKILL.md
Invoke with: /<name>

If context: fork:

Runs in: forked <agent-type> subagent

Additional Resources

• For YAML frontmatter field reference, see reference.md
• For the generation template, see templates/skill-template.md