| name | _skillwriting |
| description | how to write claude code skills |
Skill Writing Guide
Instructions
1. CRITICAL: Name First!
Folder name MUST match skill name exactly. The Skills system requires this for recognition.
- Choose skill name for precise activation (avoid accidental loading)
- Keep descriptions under 1024 characters
- Use keyword redundancy for reliable activation
2. Directory Structure
~/claude-autonomy-platform/.claude/skills/<skill-name>/
├── SKILL.md (required)
├── scripts/ (optional - for utility tools)
└── reference/ (optional - for documentation/data)
Important: Use project-level skills directory, not user-level ~/.claude/skills/
3. SKILL.md Format
---
name: exact-folder-name
description: Specific activation criteria with keywords
---
# Skill Name
## Instructions
[How to use the skill]
## Examples
[Practical examples]
4. Advanced Features
- Symlinks work: Link to central tools for maintenance
- Scripts accessible:
python3 scripts/tool.pyworks from skill context - Session-persistent: Skills stay active until session swap
- Multiple skills: Can have several active simultaneously
5. Best Practices
- Plan name and description before creating folder
- Test activation criteria carefully
- Use symlinks for central tool maintenance
- Reference docs with
[file.csv](reference/file.csv) - Avoid duplication: Cross-reference existing context instead of repeating
- Unique value only: Include only what this skill uniquely provides
- Name triggers: Put actual names/keywords you want to activate on in description
Examples
Good activation: "Calculate and record medication doses for hedgehogs"
Poor activation: "Help with animals" (too broad)
Memory skills: "Remember Amy, Erin, Ed, Orange, Delta and household members"
Name triggers: Include actual names in description for activation
Directory naming:
✅ hedgehog-dosing/ with name: hedgehog-dosing
❌ hedgehog-meds/ with name: hedgehog-dosing (mismatch!)
Cross-referencing:
✅ [See ed-care skill for protocols] (avoids duplication)
✅ [See identity doc for details] (references existing context)
❌ Repeating information available elsewhere