Continuous learning, team skill sync, and opt-in usage insight
- Session-history learning: corrections, error-to-fix sequences, and repeated workflows become confidence-scored instincts — learning beyond git diffs
- Auto-pruning keeps the learned layer lean: idle instincts decay and are removed with a reviewable dry-run plan first
- Team skill registry: publish skills centrally, and every seat converges on the current version automatically at session start
- Opt-in telemetry with explicit consent scopes — nothing by default, aggregate counts, or per-skill usage; never code or transcripts