AI Agent 错误恢复与自愈策略生成器

You are an expert AI systems reliability engineer. I need you to design a comprehensive error recovery and self-healing strategy for my AI agent system. ## System Context - Agent type: [describe your agent - e.g., coding assistant, research agent, customer service bot] - Tech stack: [e.g., Python, LangChain, OpenAI API] - Critical failure modes: [e.g., API timeout, context overflow, tool execution failure] ## Please Generate: ### 1. Error Classification Matrix Categorize potential errors into: transient, permanent, partial, and cascading failures. For each category, define severity levels (P0-P3) and expected recovery time. ### 2. Retry & Backoff Strategy - Exponential backoff configuration with jitter - Max retry limits per error type - Circuit breaker thresholds and cool-down periods ### 3. Graceful Degradation Plan For each critical capability, define: - Full capability → Degraded mode → Minimal mode → Safe shutdown - What features to disable at each level - User communication templates for each degradation level ### 4. Self-Healing Mechanisms - Automatic context window management (compression, summarization) - Model fallback chain (primary → secondary → tertiary) - Tool execution sandboxing and timeout handling - Memory/state recovery from checkpoints ### 5. Monitoring & Alerting Rules - Key metrics to track (latency, error rate, token usage, success rate) - Alert thresholds and escalation paths - Health check endpoint specifications ### 6. Implementation Code Provide Python pseudocode for the core error handling middleware that implements the above strategies. Output everything in a structured, copy-paste-ready format with code blocks.