sandbox_transcribe

Extract and transcribe up to 180 seconds from an owned audio/video file. Decode inside the sandbox, send the WAV clip to a configured audio-capable model and save a transcript file. For longer sources, process consecutive windows. Uses the selected model unless profile is supplied. Profile must declare audio input and support chat completions.

User guide / All capabilities

Contract

Effect: read. Evidence role: discovery_only.

Transcribe task audio.

Parameters

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

{
  "type": "object",
  "properties": {
    "path": {
      "type": "string"
    },
    "start_seconds": {
      "type": "number",
      "description": "Nonnegative start; default 0."
    },
    "duration_seconds": {
      "type": "number",
      "description": "1–180 seconds; default 120."
    },
    "profile": {
      "type": "string",
      "description": "Existing audio-capable model profile name."
    },
    "language": {
      "type": "string",
      "description": "Optional language hint, not a translation request."
    }
  },
  "required": [
    "path"
  ]
}

Execution requirements

The ordinary admission and tool execution checks apply.

Expected postconditions

Inspect the returned status, summary and data; do not infer success from a tool call alone.