ToolActivity

Durable and public tool activity. Raw arguments, results, queries, titles, paths, and workspace content deliberately have no field in this type. The stable target ids and bounded public error summary are the only context kept for an auditable, useful activity row.

Type object

Shape

Field	Type	Description
idrequired	ToolInvocationId	ToolInvocationId - a client-minted UUID v4, never derived from content or location.
attempt_idrequired	InferenceAttemptId	InferenceAttemptId - a client-minted UUID v4, never derived from content or location.
chat_idrequired	ChatId	ChatId - a client-minted UUID v4, never derived from content or location.
assistant_message_idrequired	MessageId	MessageId - a client-minted UUID v4, never derived from content or location.
tool_namerequired	string
read_onlyrequired	boolean
statusrequired	ToolActivityStatus
approvalrequired	ToolApproval
proposed_atrequired	Timestamp	Epoch milliseconds, UTC. Signed, so instants before 1970 are representable.
started_at	Timestamp or null
completed_at	Timestamp or null
result_bytes	integer (uint64, >= 0) or null
error_code	ToolErrorCode or null
target_workspace_id	WorkspaceId or null
target_document_id	DocumentId or null
error_message	string or null
agent_label	string or null	A local agent's own label for the call (Read README.md), bounded, kept since schema 8 so a row reads the same after the turn as during it. `None` on Koru's own tool rows.
agent_summary	string or null	What a local agent's permission request asked, bounded, kept since schema 8.
agent_input	string or null	What the agent asked the tool to do - a command line, a change's paths, an MCP call's arguments - bounded to [`crate::AGENT_CONTENT_CHARS`], kept since schema 8 (UC17).
agent_output	string or null	What the tool produced - output, a diff, a result - bounded to [`crate::AGENT_CONTENT_CHARS`], kept since schema 8 (UC17).
text_offset	integer (uint32, >= 0) or null	How much of the reply's Markdown, in bytes, had arrived when the call began: the text before it was said before it, the text after it after. Kept since schema 10 on a local agent's rows, so a reopened reply can still tell the agent's running commentary from its final answer. `None` on earlier rows and on Koru's own tool rows.
live	AgentActivityLive or null	A local agent's live state for this row - the choices a pending permission request offers - populated only from the worker's live turn table while the turn runs. `None` on every repository read and after the turn ends. The agent's wording lives on the row's own `agent_*` fields.

Example

{
  "id": "00000000-0000-4000-8000-000000000000",
  "attempt_id": "00000000-0000-4000-8000-000000000000",
  "chat_id": "00000000-0000-4000-8000-000000000000",
  "assistant_message_id": "00000000-0000-4000-8000-000000000000",
  "tool_name": "string",
  "read_only": false,
  "status": "proposed",
  "approval": "not_required",
  "proposed_at": 0,
  "started_at": 0,
  "completed_at": 0,
  "result_bytes": 0,
  "error_code": "permission_denied",
  "target_workspace_id": "00000000-0000-4000-8000-000000000000",
  "target_document_id": "00000000-0000-4000-8000-000000000000",
  "error_message": "string",
  "agent_label": "string",
  "agent_summary": "string",
  "agent_input": "string",
  "agent_output": "string",
  "text_offset": 0,
  "live": {
    "choices": [
      {
        "id": "string",
        "intent": "allow_once",
        "label": "string"
      }
    ],
    "questions": [
      {
        "id": "string",
        "header": "string",
        "question": "string",
        "options": [
          {
            "label": "string",
            "description": "string"
          }
        ],
        "allow_other": false,
        "secret": false
      }
    ]
  }
}

The contract has no example values, so example bodies are generated from each schema: every field is shown with a placeholder of its type.

Generated from contract version 1. Last updated November 12, 2026.