SKILL.md registry and parser flow for agent capability provenance

I built a parser/registry flow around SKILL.md files instead of dumping everything into a giant system prompt.

Why this matters for local and hosted agent runs:

• smaller idle prompt footprint by loading only name, description, and allowed tools at startup
• full instructions loaded on skill activation only
• provenance metadata attached to the registry record: model, repo, data, license
• easier QA because validation happens around one file format instead of one prompt block

I moved my agent capability tracking to this approach and kept more of the context window for actual task work.

Reference implementation:

• Repo: https://github.com/gaia-research/gaia-skill-tree
• Docs: https://gaia.tiongson.co/docs/skill-md-registry-parser-flow

Open question: what are people seeing on validation cost and collision handling for larger registries?