自进化Agent技能树设计提示词

You are an AI agent architect specializing in self-evolving skill systems. Design a skill tree architecture for an AI agent that learns and improves through task execution. **Agent Context:** - Agent name/purpose: [e.g., coding assistant, research agent, DevOps bot] - Primary environment: [e.g., terminal, browser, API-based] - Available tools: [list tools the agent can use] ## Design the following: ### 1. Skill Taxonomy Create a hierarchical skill tree with 3 levels: - **Domain** (top level, e.g., "Web Development") - **Capability** (mid level, e.g., "React Component Design") - **Skill** (leaf level, e.g., "Convert class components to hooks") Design at least 3 domains with 3+ capabilities each. ### 2. Skill Representation Format Define the schema for storing each skill: ```json { "id": "skill-uuid", "name": "string", "domain": "string", "capability": "string", "description": "what this skill does", "trigger_patterns": ["regex or semantic patterns that activate this skill"], "procedure": ["step 1", "step 2", "..."], "tools_required": ["list of tools needed"], "success_criteria": "how to know it worked", "confidence": 0.0-1.0, "usage_count": 0, "last_used": "timestamp", "learned_from": "task-id that created this skill", "failure_modes": ["known ways this can fail"], "refinements": ["improvements discovered over time"] } ``` ### 3. Skill Acquisition Pipeline Design the process for learning new skills from task execution: 1. **Detection**: How does the agent recognize it's doing something new? 2. **Extraction**: How does it distill the procedure from a successful task? 3. **Validation**: How does it verify the skill works reliably? 4. **Compression**: How does it generalize specific steps into reusable procedures? 5. **Integration**: How does it add the skill to the tree without conflicts? ### 4. Skill Selection Algorithm When a new task arrives, how does the agent: - Match the task to existing skills (semantic similarity + pattern matching) - Decide between using an existing skill vs. creating a new approach - Combine multiple skills for complex tasks - Handle skill conflicts (multiple skills claim to handle the same task) ### 5. Skill Evolution Rules - When does a skill's confidence increase/decrease? - When should a skill be split into sub-skills? - When should skills be merged? - When should a skill be deprecated? ### 6. Seed Skills Provide 5 starter skills (fully populated JSON) that bootstrap the agent's initial capability. Output a complete, implementable design document.