Following a task in the Agent chat

Each task the Agent works on shows as one card in the chat. The card tells you what it is doing, how far along it is and the latest steps, and holds the full list of steps under Show details.

When you ask the Agent to do something, it shows up as one card in the chat. The card stays in the same place while the Agent works, so you can read along without scrolling through every step.

While it works

The top line says what kind of task it is (Build, Research, Replies, Team task and so on), how many steps it has taken and how long it has been working. Under that is what the Agent is doing right now, in plain words, and a bar that fills as it gets through its plan. If it has no plan yet, the bar keeps moving so you can see it is still working.

The two lines under the bar are the latest two steps.

While a card is showing, the plan above the message box is hidden. The card already shows how far along the task is.

Stop is on the card while the task runs. Press it and the Agent stops after the step it is on. If it takes too long, press Force stop.

Show details

Show details opens the full list of steps. Each step is something the Agent said it would do, with the actions it took for that step listed under it. Click an action to see exactly what happened.

At the very bottom, Cost shows what the task used. Click it to see how it was charged.

If a step hit a problem and the Agent worked around it, the step is gray with a retry mark. That is normal and you don't need to do anything.

When it needs you

If the Agent needs an answer or your OK, the card turns amber and its top line says what's going on, like I can't find the 6 pricing emails or Ready to build Roofing Jobs. The question sits right under the card. Tap an answer there, or type your own, and the Agent carries on.

If more than one answer can be right, tick every option that applies and press Send. When the Agent proposes a plan for an app, the plan sits under the card: choose Build it, edit it, or say not now.

When it's done

The card folds to one line that names what got done, with the number of steps and the time it took. The Agent's reply sits under it. Open Details to see every step again.

When the Agent just answers a question and doesn't do any work, there's no card. What every task and answer used is on the Usage Dashboard in your billing settings.

If something you approved couldn't run, its line is gray and says so. The Agent carries on another way in its next message, so you don't need to do anything.

If it stops

If the task couldn't finish, the card turns red and says why in one sentence. Press Retry on the card to run it again. If the conversation has grown too long to continue, the card offers Start a new task instead.

Was this helpful?

Thanks for the signal — it helps us improve the docs.

More in Advanced

Using Overskill from AI agents (Claude, Cursor, Codex)

Connect your assistant to Overskill with the hosted MCP URL and browser sign-in. Check the connected team, access and credits before building.

Getting started with the Overskill partner API

Generate, manage, and deploy apps programmatically. Auth, the core call sequence, and working code in Node, Python, and curl.

Overskill API errors and how to handle them

What 401s, rate limits, credit exhaustion, and other API errors look like — and the right way for your code (or your agent) to respond.

Still need help?

If this didn't answer your question, our team is one click away.