A2A Task Lifecycle — A2A Protocol & Google ADK

How Tasks Flow Between Agents

In A2A, work is organized around Tasks - structured units of work that flow between a Client Agent and a Remote Agent.

Task State Machine

┌───────────┐     ┌───────────┐     ┌───────────┐
│ submitted │────▶│  working  │────▶│ completed │
└───────────┘     └─────┬─────┘     └───────────┘
                        │
                  ┌─────▼──────────┐
                  │ input-required │ (needs input from client)
                  └─────┬──────────┘
                        │
                  ┌─────▼─────────────┐
                  │ failed / canceled │
                  └───────────────────┘

Task Lifecycle Example

// 1. Client creates a task (JSON-RPC message/send or tasks/send)
POST /a2a
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "message/send",
  "params": {
    "message": {
      "role": "user",
      "parts": [{ "kind": "text", "text": "Book a flight from London to New York on 2026-05-15" }]
    }
  }
}

// 2. Remote agent processes and responds
{
  "taskId": "task_abc123",
  "status": { "state": "working" },
  "updates": [
    { "type": "status", "message": "Searching 5 airlines..." },
    { "type": "status", "message": "Found 12 flights" }
  ]
}

// 3. Agent might need clarification (input-required)
{
  "status": { "state": "input-required" },
  "question": "Do you prefer direct flights only or include layovers?",
  "options": ["direct_only", "include_layovers"]
}

// 4. Client responds, agent completes
{
  "status": { "state": "completed" },
  "artifacts": [{
    "flight": "BA177",
    "price": "$542",
    "departure": "09:15"
  }]
}

Key A2A Interaction Patterns