SkillMentor Enables LLM Agent Self-Evolution via Blind-Spot Diagnosis

SkillMentor uses reinforcement learning to train a mentor policy that identifies recurrent failure modes and curates them into corrective skills. This approach allows agents to improve via learned diagnostic capabilities rather than relying on external human supervision.