| title | Steering agents in {% data variables.copilot.copilot_cli %} | ||
|---|---|---|---|
| shortTitle | Steer agents | ||
| intro | Guide {% data variables.product.prodname_copilot_short %} during task execution to keep work on track with your intent. | ||
| product | {% data reusables.gated-features.copilot-cli %} | ||
| versions |
|
||
| contentType | how-tos | ||
| category |
|
||
| docsTeamMetrics |
|
While {% data variables.product.prodname_copilot_short %} is working on a task, you can enter a new prompt at any time. Any input you send while {% data variables.product.prodname_copilot_short %} is thinking is treated as steering and is considered in the context of the current task.
There is no separate instruction queue. To provide additional instructions, enter another prompt while {% data variables.product.prodname_copilot_short %} is running. {% data variables.product.prodname_copilot_short %} processes each message in order as part of the active task.
Steering lets you:
- Interrupt an agent that is heading in the wrong direction.
- Provide inline feedback when rejecting a tool permission request.
- Refine or clarify the task scope partway through execution.
To learn how to use {% data variables.copilot.copilot_cli_short %} to get an AI-powered review of your code changes, see AUTOTITLE.