Check reference / Token & Context Cost
TOKEN-02 Progressive disclosure
Deterministic — runs on every scan, no LLM or network needed.
Why it matters
A SKILL.md over 200 body lines with zero supporting files means all of your reference material loads into context every single time the skill fires — whether or not that invocation needs it. You pay the full token bill on every trigger.
How to fix it
Split the heavy reference sections into supporting files inside the skill directory and link to them from SKILL.md, so the model pulls them in on demand. Either a lean body (200 lines or fewer) or at least one supporting file passes this check.