skill-benchmark-gate

# Skill benchmark gate ## Contract | Field | Bound contract | |---|---| | Trigger | Before shipping any skill change; diagnosing weak skill activation, regressions, or context cost; setting release gates for skill packs | | Authority | Read and evaluate the skill; write only gate logs and issue artifacts; do not ship the change | | Side effect | Runs evaluation loops; writes run-log entries; opens follow-up issues on GitHub or equivalent tracker | | Done | A reproducible PASS or FAIL record with case results, deltas, context cost, and failure issues | ## Inputs All inputs are required and must be explicit. No defaults are inferred from the procedure. 1. Changed skill and comparison revision: the skill slug, the diff or branch under test, and the baseline revision to compare against. 2. Explicit universal activation case set: the complete list of cases that must all show activation greater than 0% with the skill enabled. Every case must be named; no case is implied. 3. Explicit critical regression case set and baseline: the list of critical scenarios and their recorded baseline results (pass/fail and context token count per scenario) from the comparison revision. Every scenario must be named with its baseline; no scenario is implied. 4. Explicit context-budget ceiling: the maximum total context tokens the skill may load, stated as a number. 5. Explicit iteration budget: the maximum number of evaluation iterations the gate may run, stated as a number. ## Procedure 1. Val...