Use when designing a fine-grained, hook-driven learning system that observes every tool call (not just session end), stores atomic "instincts" with confidence scores, and evolves clusters of related instincts into skills, commands, or agents — with project-scoped vs. global separation so React conventions don't leak into a Python repo. Trigger phrases include "observe every tool call for patterns", "score confidence on a learned behavior", "keep this pattern scoped to this project", or "cluster these instincts into a skill". For the simpler session-end, whole-skill approach, see continuous-learning.
