Knowledge Share
Technical articles, tutorials, and insights
My AI Agent Deleted My Skills and Thought It Did a Good Job
HermesAgent silently deleted my carefully tuned tech podcast production Skill and merged six other media Skills into one low-quality replacement — with a plain rm -rf, no confirmation, no rollback. The backup recovered the content, but the incident exposes a clear engineering gap: without an evaluation system, agent self-evolution is random destruction with good intentions.
Making AI Skills Better Over Time: Evaluation, Tuning, and Self-Evolution System Design
LLM self-evaluation accuracy is only 46.4% — equivalent to random guessing. This is an empirical finding from Microsoft Research and Fudan University. The three dimensions that actually matter for Skill quality are failure path encoding, executable specificity, and dangerous operation blacklists. This article explains how to design an AI Skill lifecycle management system based on the latest research (SkillLens/SkillOpt).
Open Source Project of the Day (Part 10): AgentEvolver - Efficient Self-Evolving Agent System for Autonomous Learning and Evolution
A deep dive into AgentEvolver, an open-source self-evolving agent system with 1.1k+ Stars — using three core mechanisms (Self-Questioning, Self-Navigating, Self-Attributing) to enable AI agents to autonomously generate tasks, guide exploration with experience, and achieve fine-grained credit assignment for efficient self-evolution