Inbox
The inbox surfaces items the agent has explicitly raised to you. If it’s empty, no agent has called ask_user_question or asked you to complete an external action — that’s normal.
The five item kinds
Section titled “The five item kinds”input_required — the agent called ask_user_question
Section titled “input_required — the agent called ask_user_question”Up to 5 multiple-choice questions; you submit answers. Example:
“Which Stripe environment should I target?” [Live] [Test] [Other…]
You can also answer by simply typing a reply in the chat session (web and Telegram) — a plain text message while the question is pending is consumed as the answer; an exact choice-label match selects that choice, anything else is recorded as “Other”. Messages with images or attachments don’t count as answers, and neither does deleting the item — the session stays blocked. If you don’t answer within 30 minutes the agent gets a cancelled signal and decides what to do next; after that, a typed reply is delivered as an ordinary message rather than an answer.
action_required — the agent needs you to do something outside the chat
Section titled “action_required — the agent needs you to do something outside the chat”Browser login walls, MFA, CAPTCHA, OAuth approval, file pickers, manual web actions. Open the session to interact with the agent’s browser; click I completed this when done. The session wakes up.
task_complete — long-running session finished
Section titled “task_complete — long-running session finished”Informational. Producers: /goal, /mission, /loop. Acknowledge or open to inspect.
progress_checkin — long-running session update
Section titled “progress_checkin — long-running session update”Pure status update mid-run. No action required.
Off unless the session was created with a check-in interval — nothing in Studio sets one, so in practice these appear only on sessions started through the API.
governance_gate — the agent needs your approval for a tool call
Section titled “governance_gate — the agent needs your approval for a tool call”Raised when the agent calls a tool your governance policy marks as approval-required. The item names the tool, the reason it was blocked, and an excerpt of the call’s arguments, with Approve / Reject.
Either decision is posted back into the session, so the agent picks it up straight away. See Tools that need approval first for what an approval covers and how long it lasts.
Statuses
Section titled “Statuses”pending → acknowledged / responded / expired (terminal). Default views hide expired.
Triage order
Section titled “Triage order”input_required → action_required → governance_gate → informational items. The first two block the agent outright; an unanswered governance_gate leaves it without a tool it asked for; the rest are informational.