Creating tasks
How to create prompt and command tasks, assign them to desktops, and run them through Mission Control.
Task types
Mission Control supports two types of tasks:
Prompt tasks send a natural language instruction to your AI agent. The agent interprets the prompt, plans its approach, and executes autonomously. Use these for open-ended work like "set up a Node.js project with Express and write unit tests" or "research competitors and write a summary."
Command tasks run a shell command directly on the desktop. The command runs as-is, with no AI interpretation. Use these for precise operations like npm install, git pull, or restarting a service.
Creating a task from the dashboard
- Open Mission Control and click New Task
- Pick the task type: Prompt or Command
- Enter the task content, either your prompt text or the shell command
- Select the target desktop from the dropdown (only running desktops appear)
- Optionally toggle Requires Approval if you want to review the result before it is marked done
- Click Create
The task appears in the Pending column of the Kanban board.
Running a task
Creating a task does not start it automatically. This gives you time to review, batch tasks, or wait for the right moment.
To run a pending task, click the task card on the Kanban board and select Run. Programmatic task execution is not exposed in the public API in this release.
Once running, prompt tasks go to the desktop's AI agent, while command tasks run directly on the desktop's shell.
Task fields
| Field | Type | Required | Description |
|---|---|---|---|
type | "prompt" or "command" | Yes | How the task is executed |
content | string | Yes | The prompt text or shell command |
desktopId | string | Yes | Target desktop ID |
requiresApproval | boolean | No | If true, task enters "review" status instead of "done" |
Creating tasks programmatically
Programmatic task creation is not exposed in the public API in this release. Create and run tasks from the dashboard.
Batch workflows
You can create several tasks targeting different desktops from the dashboard and then run them together or one at a time:
- Create all tasks from Mission Control (they start in "pending")
- Run them one by one, or fire them all at once
- Watch progress on the Kanban board or in the activity stream
If approval gates are on, each task pauses at "review" so you can check the result before it finalizes. See Approval workflow for details.
Tips
- Prompt tasks work well for open-ended work where the agent needs to figure out the approach.
- Command tasks work well for deterministic operations where you know the exact command.
- Specific prompts get better results. "Install nginx and configure a reverse proxy for port 3000" works better than "set up the server."
- For long-running commands, check the activity stream for real-time progress.