Questions, runs, and replays
Answer a bot's multiple-choice questions, read the line under a finished reply, replay or share a run, and save a run that worked as a reusable skill.
Questions from a bot
Sometimes a bot needs you to make a choice before it can carry on. It shows a question card: a short heading, the question, and lettered options. If the bot didn't give a heading, the card says The agent has a question.
Report format Which format do you want for the report? A. PDF B. Word document C. Markdown D. Skip
To answer:
- Click an option.
- Or type your own answer in Type your own answer and press Enter.
- Or click Skip to let the bot decide.
- Clicking × in the corner dismisses the question.
If Machine Desktop is in the background, you get a notification: "Scout has a question". The bot waits for you.
If you see "The agent's question could not be reached to answer it", just type your answer as a normal message instead.
Want fewer questions? Put your preferences in the bot's Soul, like "Always deliver reports as PDF." Or, to get more questions, add "Ask me before you start anything long."
The run footer
When a bot finishes a reply that took real work, a quiet line appears under it:
Worked 1m 05s · 6 steps · $0.04 · Replay · Share
| Part | Means |
|---|---|
| Worked … | How long the turn took, from your message to the reply. |
| … steps | How many actions the bot took, like searches, commands, and file changes. |
| The cost | What the turn cost, when there was a cost. |
The line doesn't name the model. To see which model a thread runs on, look at the model chip in the chat header.
There's no line for a plain reply where the bot didn't need to do anything.
Replay
Click Replay to open the Live view, the bot's computer shown as a desktop, where you can watch the steps it took. See Replay a turn.
Share
Click Share to save the run as one web page anyone can open in a browser, like scout-replay.html. It downloads to your computer, with the screenshots inside it, so you can send the file to a colleague. It contains what the bot did in that turn, so check it before sending it on.
The run card
When a bot ran several commands to complete your request, a small run card appears beside the message box after the turn finishes. While the bot is still working, the live work card in the chat already shows what it's doing.
The run card starts folded: a slim pill with a summary like 3 steps · 1 failed. Click Expand the run (the arrow) to see each step, under the heading This run:
| Mark | Meaning |
|---|---|
| ✓ | The step worked. |
| ✗ | The step failed. |
Steps may also be tagged verified (the bot checked the result) or dry run (a test that didn't change anything).
- Click Collapse the run to fold it again.
- Click × (Dismiss the run) to close it. It comes back the next time the bot runs a command.
A failed step doesn't always mean the task failed. Bots often try something, see it fail, and fix it. Read the bot's reply for the outcome.
Save a run as a skill
When a bot does something well that you'll want again, like "pull last week's sales numbers and make a chart", turn it into a skill: a saved recipe the bot can reuse in future threads.
Click Save as skill
On the run card, click Save as skill. It's there when at least one step worked and the bot isn't busy. It's on the folded pill too.
Add notes
The run and your original request are added to your message box. Nothing is sent yet. Add anything the skill should always do, like "Always use Monday to Sunday" or "Save the chart as sales-chart.png."
Send
The bot drafts the skill. Check it in Bot settings → Skills: click its name to read it.
You can also draft a skill at any time by sending /learn. See Skills and
memory.
To stop bots from drafting skills, turn off Settings → Experimental → Bots may draft skills for your review. Save as skill then disappears.