skillify

[!IMPORTANT] AI Context & Knowledge Heritage

• Subsystem: Agent Skills Registry / skillify
• Architecture: @docs ARCHITECTURE:Documentation
• Failure Path: Information drift, legacy terminology, or documentation mismatch.
• Observability: Traceability via execution/parity_guard.py ([SKILL])

Skillify — Auto-Create Skills from Workflows

Turn repetitive patterns into reusable skills. If you've done it three times, it should be a skill.

When to Skillify

✅ Good candidates:

• You've seen the user ask for the same type of work 3+ times
• A workflow involves 5+ consistent steps
• The pattern works across different projects
• Other agents could benefit from this knowledge

❌ Bad candidates:

• One-off tasks (just do them)
• Project-specific hacks (use memory instead)
• Already covered by existing skills (check first)

---

Skill Creation Protocol

Step 1: Identify the Pattern

What triggers this workflow? (user says X, file type Y, domain Z)
What steps are always the same?
What parts vary between uses?
What's the expected output?

Step 2: Generate SKILL.md

Skill Authoring Rules (L1/L2/L3 Progressive Disclosure Specification):

1. L1 (Metadata): name MUST be kebab-case ($\le 64$ chars). description MUST be trigger-optimized with specific action keywords ($\le 1024$ chars).
2. L2 (Instructions): Keep SKILL.md body under 5,000 words focusing purely on procedural instructions. Use bold emphasis for non-negotiable assertions.
3. L3 (Resources): Offload heavy API schemas or docs to references/, boilerplate to assets/, and fragile execution logic to deterministic scripts in scripts/.
4. Checkable Completion Criteria: Every step MUST end on an explicit, checkable criterion.

Use this template:

---
name: [kebab-case-name-max-64-chars]
description: [Actionable, trigger-optimized summary describing when to activate this skill, max 1024 chars]
when_to_use: "[When the user asks X, works with Y files, or Z domain. NOT for A.]"
allowed-tools: [Read, Write, Edit, Grep, Glob, Bash — only what's needed]
disable-model-invocation: true # Set to true if manually invoked only
effort: [low | medium | high]
---

# [Skill Name] — [Short Subtitle]

> [One-line philosophy or principle]

## Overview
[2-3 sentences explaining what this skill enables]

## When to Use
✅ Good for: [list]
❌ Not for: [list]

## Protocol
### Step 1: [First Action]
[Action-first instructions with **bold emphasis** on strict constraints]

### Step 2: [Second Action]
[Instructions]

### Step N: [Verification]
[Explicit, checkable completion criterion]

## Best Practices
[3-5 key rules]