AI Agent 自进化技能树种子设计器

You are an expert AI agent architect specializing in self-evolving skill tree systems. I want to design a **self-evolving agent framework** with these requirements: 1. **Seed Design**: Start from a minimal seed (under 5K lines) that contains only core capabilities: file I/O, shell execution, and a skill registry. 2. **Skill Tree Growth**: The agent should automatically: - Identify recurring task patterns - Extract reusable skills from successful task completions - Store skills as composable, versioned modules - Build dependency graphs between skills 3. **Token Optimization**: Design the context injection strategy to achieve at least 5x token reduction by: - Loading only relevant skills per task - Compressing skill descriptions into minimal signatures - Using a retrieval-based skill selector instead of stuffing all skills into context 4. **Self-Improvement Loop**: - After each task, evaluate: Was a new skill learned? Was an existing skill improved? - Periodically prune underused skills - Merge similar skills into generalized versions Please provide: - The seed architecture (modules, interfaces, data flow) - Skill schema (name, trigger conditions, implementation, dependencies) - Growth algorithm pseudocode - Token budget analysis comparing naive vs. skill-tree approach Output in structured markdown with code blocks where appropriate.