llms: Skill for Claude Code

Skillify

Converts a working solution into a permanent skill with routing tests, registry entry, and DRY compliance. The complement of /hookify (which blocks failures) — this crystallises successes.

Usage

/skillify <name>              # Full 8-step pipeline for a new skill
/skillify <name> --check      # Audit an existing skill (steps 3-8 only)
/skillify <name> --step 3     # Run a single step

When to use

• You just solved something ad-hoc and want it reusable
• A workaround worked 3+ times → time to make it a skill
• /memory extract surfaces a recurring pattern worth formalising
• After /quick or a raw session where something clicked

8-Step Pipeline

Run steps in order. Stop and report if any step fails — do not skip.

---

Step 1 — SKILL.md contract

If ~/.claude/skills/<name>/SKILL.md does not exist, create it.

Required frontmatter:

---
name: <name>
description: "<one sentence, < 200 chars, no quotes inside>"
allowed-tools: Read, Write, Edit, Bash, Grep, Glob # trim to what's needed
---

If it exists, validate:

• name: matches directory name
• description: present and < 200 characters
• allowed-tools: present

Verify: parse frontmatter with Python — must not raise yaml.YAMLError.

python3 -c "
import yaml, sys
fm = open('$HOME/.claude/skills/<name>/SKILL.md').read()
end = fm.find('---', 3)
d = yaml.safe_load(fm[3:end])
assert 'name' in d and 'description' in d, 'missing fields'
assert len(d['description']) < 200, f'description too long: {len(d[\"description\"])}'
print('OK:', d['name'], '-', d['description'][:60])
"

---

Step 2 — Deterministic extraction

Read the SKILL.md body. Identify logic that is:

• Computing / parsing / transforming data (same input → same output)
• Making API calls and parsing structured responses
• Searching files or running commands and collecting results

If found: create ~/.claude/skills/<name>/scripts/<name>.py (or .sh) and replace the prose description with a Run: block that calls the script.