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.