AI Agent 异步任务编排与回调设计模板

You are a senior software architect specializing in AI agent systems and distributed task orchestration.

I need you to design an async task orchestration system for my AI agent.

Agent Description:

• Agent purpose: [What does your agent do?]
• Steps involved: [List the main steps, e.g., 1. Web search 2. Extract data 3. Analyze 4. Generate report]
• External APIs used: [e.g., search API, database, LLM calls]
• Expected execution time: [e.g., 30s-5min per run]

Please design:

1. State Machine: Define all states (pending, running, waiting_callback, retry, failed, completed) with transition rules
2. Task DAG: A directed acyclic graph showing task dependencies and parallelization opportunities
3. Retry Policy: Per-step retry config with exponential backoff, max attempts, and circuit breaker thresholds
4. Timeout Handling: Cascading timeouts (step-level, pipeline-level, global) with graceful degradation
5. Callback Design: Webhook/polling patterns for long-running steps, with idempotency keys
6. Error Recovery: Checkpoint/resume strategy so failed runs can continue from last successful step
7. Observability: Structured logging schema, trace IDs, and metrics to emit

Provide the design as:

• A Mermaid state diagram
• TypeScript/Python interface definitions for the task queue
• Example retry/timeout configuration JSON
• A sample error recovery flow