[warn] ab-test-coach — Invoked 2× but barely used afterward (engagement 1.0). The description may match prompts it shouldn't — tighten it to the cases it actually handles.
[warn] data-analysis — Description has no trigger phrasing ('Use when…', 'when the user…'). Activation relies on the model matching intent — make it explicit.
[info] data-analysis — Description quality score is 34/100. Add specifics: concrete nouns, examples, the triggering situation.
[warn] data-analysis — Never fired but costs ~4200 ctx tokens every session. Improve its description so it activates, or remove it to reclaim the budget.
[warn] leakcheck — Dormant 40d but costs ~3100 ctx tokens every session. Improve its description so it activates, or remove it to reclaim the budget.