AI Agent 技能树自进化与基因编程框架设计

You are a researcher specializing in self-evolving AI agent architectures. Design a Gene Expression Programming (GEP) inspired framework for AI agent skill evolution.

Core Concept

AI agents should not be static — they should evolve their skill trees through:

1. Skill Genome: Each skill is encoded as a gene (tool definition + usage pattern + success metrics)
2. Mutation: New skills emerge from combining/modifying existing ones
3. Selection: Skills that solve tasks efficiently survive; inefficient ones are pruned
4. Expression: The agent active skill set adapts to the current task environment

Design Requirements

Skill Genome Schema

Define a JSON schema with: skill_id, name, description, trigger_conditions, tool_chain, fitness_score, generation, parent_skills, mutation_log

Evolution Loop

1. Agent encounters a task it cannot solve efficiently
2. Analyze failure patterns and identify skill gaps
3. Generate candidate skill mutations (combine existing skills, add new tool usage patterns)
4. Test candidates in sandbox environment
5. Promote successful mutations to active skill tree
6. Archive deprecated skills with reasoning

Output

Provide:

• Complete skill genome specification
• Evolution algorithm pseudocode
• Fitness evaluation criteria (success rate, token efficiency, time to completion)
• Skill tree visualization format (Mermaid)
• Safety constraints (prevent dangerous skill evolution)
• Example: evolve a web scraping skill into a structured data extraction skill

Target Agent

Agent type: [AGENT_TYPE] Current skills: [LIST_CURRENT_SKILLS] Target domain: [DOMAIN]

Begin with the theoretical framework, then provide implementation details.