Type: technique
Status: draft
Domain: horizontal
Tags: knowledge, ai-pm, technique, agents, composability, skill-creation
Last updated: 2026-02-14

Meta-Skill Pattern

Summary

A technique for bootstrapping agent capabilities by building a “skill that builds skills” — a meta-skill that takes a description of a desired workflow and generates a properly structured skill folder (instruction file, templates, validation scripts). Instead of hand-authoring each new skill from scratch, you create one factory skill and then use it to produce domain-specific skills consistently and quickly.

The underlying insight is composability: agent skills aren’t just endpoints — they can invoke and build on each other. A meta-skill is the simplest expression of this, but the pattern extends to any skill that orchestrates or generates other skills.

How to Apply

When to use: When you need to create multiple skills and want consistency in structure, metadata, and quality. Especially useful when onboarding a team to skill-based workflows — the meta-skill encodes your conventions so each team member produces well-formed skills without memorizing the spec.

When not to use: For a single, one-off skill, the overhead of building a meta-skill isn’t justified. Just create the skill directly.

Steps:

  1. Create a create_skill (or similar) folder containing a SKILL.md that instructs the agent how to generate a new skill — including folder structure conventions, YAML metadata requirements, and template files to include
  2. Optionally include a validation script (e.g., validate_skill.py) that checks YAML formatting, file references, and content structure. This bundles quality assurance into the generation process itself
  3. Open your skills directory in an AI-native editor or start Claude Code in that directory — the agent auto-discovers the meta-skill from the filesystem
  4. Prompt the agent: “Use the create_skill skill to create a skill for [your workflow]”
  5. The agent reads the meta-skill instructions, generates the new skill folder, and (if validation is bundled) runs the validation script automatically
  6. Review the generated skill, adjust the instruction file for tone/format, and iterate

For AI PMs: This pattern is relevant beyond personal productivity. It demonstrates how to scale agent capability creation across a team: instead of relying on one person to author all skills, you distribute a meta-skill that encodes your quality standards. The meta-skill becomes an onboarding tool — new team members can generate skills that meet your conventions from day one.

Sources

From: 2026-02-14 Claude Skills Explained Reusable AI Agents

Key quote: “Instead of relying on Claude’s built-in creator, I decided to build my own. I started a chat with Cursor’s AI and gave it the official Anthropic Skills Documentation as context.” Attribution: Claire Vo What this source adds: Vo demonstrates the full cycle: building a meta-skill in Cursor (~3 minutes), then using it via Claude Code to generate a changelog-to-newsletter skill. The meta-skill auto-generated a Python validation script alongside the instruction file — showing that skills can bundle executable tooling, not just natural language instructions. The article also surfaces a practical contrast: Claude’s built-in skill creator over-generates (12 files when 5 suffice), while a hand-crafted meta-skill produces exactly what you specify. Links: Original | Archive


This site uses Just the Docs, a documentation theme for Jekyll.