Default output: return only the result, blockers, and required evidence. Omit preambles, process narration, repeated context, confidence scores, and follow-up offers. Use at most five bullets unless a required artifact or schema needs more.

Sync skill taxonomy

Keeps the skill catalog consistent: every skills/**/SKILL.md carries a metadata.category (user-invoked or model-invoked), and every user-invoked skill carries disable-model-invocation: true so its description is excluded from model routing context.

Why

Per the Agent Skills spec, every skill's name + description loads at startup for routing (~100 tokens each). disable-model-invocation: true (a Claude Code / Cursor / Pi extension, not part of the base spec) marks a skill as slash/manual-only — its description is NOT loaded for routing, saving tokens and preventing the model from auto-invoking skills meant for explicit triggers.

Drift happens: new skills land without the flag, descriptions get rewritten, trigger semantics shift. This skill re-applies the taxonomy deterministically.

When to run

• After adding, renaming, or removing skills
• After rewriting a skill's description
• When reviewing token cost of the skill catalog at startup
• When routing quality degrades (model invoking skills it shouldn't)