Context editing response

You can see which context edits were applied to your request using the context_management response field, along with helpful statistics about the content and input tokens cleared.

{
  "id": "msg_013Zva2CMHLNnXjNJJKqJ2EF",
  "type": "message",
  "role": "assistant",
  "content": [
    // ...
  ],
  "usage": {
    // ...
  },
  "context_management": {
    "applied_edits": [
      // When using `clear_thinking_20251015`
      {
        "type": "clear_thinking_20251015",
        "cleared_thinking_turns": 3,
        "cleared_input_tokens": 15000
      },
      // When using `clear_tool_uses_20250919`
      {
        "type": "clear_tool_uses_20250919",
        "cleared_tool_uses": 8,
        "cleared_input_tokens": 50000
      }
    ]
  }
}

For streaming responses, the context edits are included in the final message_delta event:

{
  "type": "message_delta",
  "delta": {
    "stop_reason": "end_turn",
    "stop_sequence": null
  },
  "usage": {
    "output_tokens": 1024
  },
  "context_management": {
    "applied_edits": [
      // ...
    ]
  }
}