start_long_task

Optionally declare sustained research to retain a larger evidence slice. Ordinary tasks already have no default tool-call cap. This does not change permissions or an explicit operator limit. Only one declaration per turn is needed.

User guide / All capabilities

Contract

Effect: read. Evidence role: discovery_only.

Inert runtime-control capability. Runtime intercepts this tool name before dispatch to prepare evidence context; this run() is never actually invoked.

Parameters

This is the current provider-facing JSON Schema, generated from the runtime definition.

{
  "type": "object",
  "properties": {
    "reason": {
      "type": "string",
      "description": "Short human-readable justification. Shown to the user in the progress status line and written to runtime trace. Example: 'paginar 2000 mensagens do @one_person em 10 canais'."
    }
  },
  "required": [
    "reason"
  ]
}

Execution requirements

The ordinary admission and tool execution checks apply.

Expected postconditions

• per-turn maxEvidenceSlice is raised to settings.runtime.longTask.evidenceSliceFloor