edit-skill

# Instructions for Creating and Modifying Skills

## MANDATORY Validation (CREATE only)

**STOP**: Before proceeding, you MUST explicitly answer these questions:

### Complexity Evaluation Checklist

Answer each question with YES or NO:

**Q1: Will this be auto-invoked frequently enough to justify context pollution?**
- Answer: [YOU MUST STATE YES OR NO]

**Q2: Can this be kept under 500 tokens (concise capability)?**
- Answer: [YOU MUST STATE YES OR NO]

**Q3: Is this a specific capability rather than a workflow?**
- Answer: [YOU MUST STATE YES OR NO]

### Decision Tree

**If ANY answer is NO:**
→ STOP. Do NOT create a skill.
→ Explain why it's not appropriate.
→ Recommend one of these alternatives:
  - **Direct request** - For simple one-off tasks (just ask Claude)
  - **Slash command** - For user-initiated workflows or verbose instructions
  - **Sub-Agent** - For complex multi-step exploration/research tasks
→ EXIT this skill immediately.

**If ALL answers are YES:**
→ Proceed to "Creating New Skills" section below.

---

## Uncertain About Tool Type?

If unsure whether this should be a skill vs command vs agent vs script, see the `edit-tool` orchestrator for comprehensive decision matrix.

**Quick checks:**
- Token budget >500? → Probably not a skill
- Used <5 times per session? → Probably command
- Deterministic shell-only? → Probably bash script
- Complex exploration? → Probably sub-agent

## Best Practices Reference

For latest Anthropic skill authoring guidelines, use WebSearch:
```
WebSearch: "skill authoring best practices site:platform.claude.com"
```

See `reference.md` for detailed decision frameworks and patterns.

## Determine Action Type

**CREATE**: New skill requested
**MODIFY**: Update existing skill (keywords: "update", "modify", "improve", "add to", "fix")

## Creating New Skills

1. Ask for details if missing: purpose, triggers, tools needed, model selection
2. **Location**: Use `.claude/skills/` for project skills or plugin `skills/` directory