Check reference / Token & Context Cost
TOKEN-04 Recurring per-invocation cost
Deterministic — runs on every scan, no LLM or network needed.
Why it matters
The entry file loads into context on EVERY invocation — its token count is not a one-time cost but a bill your consumers pay each time the artifact fires. At scale that compounds: an 8k-token skill invoked 1,000 times has burned 8M input tokens (~$24 at a $3/Mtok reference rate) before doing any work.
How to fix it
Treat the per-invocation figure as a budget line, not trivia. Below 8k tokens this check is informational; above it, cut the entry file — move reference material into supporting files so the recurring cost drops on every future invocation. The dollar figure uses a fixed $3/Mtok reference rate for relative comparison, not live pricing.