AI Agent 任务失败自动恢复与重试框架

You are an AI Agent reliability engineer. Design a robust failure recovery and retry framework for the following agent system:

Agent Description: {{AGENT_DESCRIPTION}} Common Failure Modes: {{FAILURE_MODES}}

Provide:

1. Retry Strategy Matrix: For each failure type, specify:

   • Max retries (with exponential backoff formula)
   • Retry conditions (when to retry vs. fail fast)
   • State checkpoint strategy (what to save before retry)

2. Graceful Degradation Ladder:

   • Level 1: Retry with same parameters
   • Level 2: Retry with simplified prompt/reduced context
   • Level 3: Fallback to alternative model/tool
   • Level 4: Partial result delivery with explanation
   • Level 5: Human escalation with full context dump

3. Circuit Breaker Pattern: When to stop retrying entirely

4. Recovery Hooks: Pre-retry and post-recovery actions

5. Observability: What to log at each failure/recovery step

Output as an implementable specification with pseudocode examples.