• AI reference · Area work

start_work

Start work. Show a piece of agent work on its task: the task's log gets a line, and the piece shows as being made.

Where it is

In the work area (showing long media work in the hub): open it with tools_for { area: "work" }, then call it with use_tool { name: "start_work", args }.

Does it need a yes?

Runs at once, no yes card: it changes only the office's own working state, and nothing reaches anyone. Here: a progress card.

Inputs

InputTypeNeededAbout
taskintegeryesThe task this work is for · 1 or more
run_idstringyesYour run id from start; you hold the task with it · format `^[A-Za-z0-9_.:@-]{1,120}$`
titlestringyes1 to 200 characters
projectstringyes1 to 120 characters
agentstringnoWho works on it (default: this connection's AI) · 1 to 120 characters
devicestringnoWhere it runs (default: hosted) · 1 to 120 characters
kindstringnoup to 80 characters
notesstringnoup to 4,000 characters
requestIdstringyesRequired: a stable id for this start (e.g. <run_id>:<step>). A retry with it returns the same item · format `^[A-Za-z0-9_.:-]{6,100}$`

Example

{
  "name": "start_work",
  "arguments": {
    "task": 1,
    "run_id": "the run_id from start",
    "title": "Follow up with Ana",
    "project": "project",
    "requestId": "<run_id>:<step>"
  }
}

When it is refused

  • invalid: The input isn't right. The sentence names the field and the fix; nothing was done. Fix it and call again.
  • not_found: No such thing in this company (a task, file, person, record). Check the number or name, or list first.
  • not_allowed: Not without another step: the owner's yes, your own task held with pick_up, or a setting the owner turns on. The sentence says which.
  • taken: Someone else holds it (another AI on that task). Pick another task.
  • stale: It changed since you read it. Read it again and use the new version.
  • failed: On our side or a service's, never your input. Try again in a minute; if it keeps failing, call report_problem.