self-learning-loop

Skill for continuous self-learning of an agent through an install -> evaluate -> refine cycle: selects 1–2 most useful improvements, checks metrics, and commits only those that provide gains or preserve stability.