write-a-skill

name	write-a-skill
description	Create new agent skills with proper structure, progressive disclosure, and bundled resources. Use when user wants to create, write, or build a new skill.

Writing Skills

I write skills as behavior interventions — each one solves one named failure mode. Skills are organized into buckets (engineering/productivity/misc/personal), public skills indexed in plugin.json and README, personal skills kept local.

Process

1. Gather requirements — ask about:

   • What failure mode does this skill prevent?
   • What specific use cases should it fire on?
   • Does it need executable scripts or just instructions?
   • Any reference materials or real examples to include?

2. Draft the skill — create:

   • SKILL.md with terse instructions (target: <150 lines for most skills)
   • Trigger phrase in the YAML description ("Use when…")
   • One explicit anti-patterns section (failures I've seen)
   • Additional reference files only if SKILL.md would exceed 100 lines

3. Review with user — present draft and ask:

   • Does this capture the failure mode?
   • Anything missing or unclear?
   • Should any section be more/less detailed?

Skill Structure

skill-name/
├── SKILL.md              # Main instructions (required)
├── REFERENCE.md          # Detailed docs (if needed)
└── scripts/              # Utility scripts (if needed)
    └── helper.js

SKILL.md Template

---
name: skill-name
description: One-sentence failure mode. Use when [specific triggers].
---

# Skill Name

[Opening: what this skill prevents, in first-person voice]

## When this skill is active

[Describe the contexts where this fires]

## Phases / Checklist

[ ] Phase 1
[ ] Phase 2

## Anti-patterns

- **Bad pattern X** — why it fails
- **Bad pattern Y** — consequences

Key Guidelines

• Voice: first-person for public skills ("I do X", "in my code"), pragmatic tone
• Description: max 1024 chars, concrete triggers ("Use when user says 'foo'")
• Length: aim for <150 lines — if longer, split into REFERENCE.md
• Anti-patterns: ≥3 explicit callouts of what agents do wrong
• Examples: 1-2 concrete code snippets or scenarios
• Structure: for complex skills, use phases or checklists with explicit breakpoints

When to Add Scripts

Add utility scripts when:

• Operation is deterministic (validation, formatting)
• Same code would be generated repeatedly
• Errors need explicit handling

When to Split Files

Split into separate files when:

• SKILL.md exceeds 100 lines
• Content has distinct domains (separate testing strategies)
• Advanced features are rarely needed

Bucket Assignment

• engineering/ — solves daily code work failures
• productivity/ — solves workflow/meta failures
• misc/ — kept around, rarely used
• personal/ — tied to your projects, not promoted
• in-progress/ — drafts, not ready
• deprecated/ — no longer used

Public skills (engineering/productivity/misc) must appear in:

• Top-level README.md Reference section
• .claude-plugin/plugin.json
• Corresponding bucket README.md

Review Checklist

• One clear failure mode named in opening
• Trigger phrase in description ("Use when...")
• SKILL.md under 150 lines (split if needed)
• ≥3 anti-patterns explicit
• ≥1 concrete example
• No time-sensitive info
• First-person voice for public skills
• Bucket assignment correct (public ⟹ plugin.json+README)