Usage & Features
Library Health
On this page
Library Health tracks which skills your agents actually load and surfaces safe cleanup recommendations — similar to a storage scan that highlights files you rarely use.
What it tracks
- Skill reads via Cursor
postToolUsehooks when agents openSKILL.mdfiles chain captureevents (secondary signal, deduped)- Manual test events via
chain skill-usage touch <slug>
All data stays local in brain.db under your hub folder. Nothing is sent to the network.
Enable tracking hooks
Automatic (recommended): chain setup merges active-project and Library Health hooks into ~/.cursor/hooks.json when Cursor is detected. You can also run:
Or use Skills → Library Health → Install Cursor hooks in the hub.
Manual: merge entries from core/hooks/cursor-hooks.example.json into ~/.cursor/hooks.json (user-global) or a project .cursor/hooks.json.
Hub views
- Dashboard → Library Health — summary card with dormant count and reclaimable size
- Skills → Library Health (
#skills/insights) — full list, filters, and batch Deactivate (reversible)
Recommendations stay review-first. Core skills are never suggested for removal. Batch deactivate requires explicit confirmation and supports at most 20 skills per action.
CLI
Disable tracking in Hub Settings (skillLibrary.trackingEnabled in hub-settings.json) if you prefer not to record invocations. See Advanced settings for other hub preference blocks.