Before your AI writes a skill, it reads the masters

What it does

This is a meta-skill for Claude Code, Codex, and similar agent platforms. When you ask to create a new skill—say, for user interviews or writing PRDs—it pauses generation to research established methodologies from domain experts. It checks a local database spanning 15+ domains, searches for additional experts, finds exemplary outputs, catalogs common mistakes, and cross-validates across sources. Only then does it hand off to a separate skill-creator tool to generate the final skill.

The interesting bit

The core insight is that most AI skills are just vibes dressed up as instructions. This project treats expertise curation as the hard part, not syntax formatting. It even maintains an “Oral Tradition” section for experts who publish through talks rather than books, acknowledging that not all mastery lives in documentation.

Key highlights

• Curated methodology database covering product, sales, engineering, negotiation, and more, citing names like Marty Cagan, Barbara Minto, and Chris Voss
• Three-layer search: local database, web expert discovery, then deep dives into primary sources
• Explicitly hunts for anti-patterns and golden examples to set a quality bar
• Includes two secondary skills: one searches trusted skill marketplaces instead of building from scratch, and another distills knowledge from high-quality GitHub projects into portable skills
• Quality checklist enforces cross-validation and primary-source fidelity before finalizing anything

Caveats

• Requires the external skill-creator skill to actually produce the final output; this repo is the research phase only

Verdict

Grab this if you use Claude Code or similar agents and are tired of generating shallow, generic skills. Skip it if you are looking for a standalone skill generator or a traditional software library—this is a curation layer, not a code framework.