Input and Context

Use the Composer to describe tasks, provide relevant materials, reference existing content, or invoke commands. While a task is running, you can also send follow-up requirements, adjust direction, or compact the context.

Provide context

When describing a task, state the goal first, then add relevant information—this usually helps the Agent understand the task faster.

Entry	How to trigger	What it does
Add attachments	Paste, drag, or click +	Upload screenshots, documents, images, videos, or other materials as context for the current task
Insert @ mentions	Type @	Reference files or folders in the workspace so the Agent can locate the relevant code or material
Insert / commands	Type /	Invoke built-in commands or saved extension commands to reuse fixed workflows
Screenshot	Click + and choose Screenshot, or use the shortcut macOS: Shift + Cmd + S Windows: Shift + Ctrl + S	Select a screen region and add comments to help the Agent locate a page position or error message; grant the permission as prompted on first use
Comment and add to conversation	Select content, then choose Comment or Add to conversation	Quote text, code, or tool results from the conversation into the current task, making it easier to follow up or adjust direction

You can also provide error logs, reproduction steps, expected behavior, verification commands, and the directories, files, or behavior that must not be changed.

Screenshots capture any region of the screen. To annotate a specific element or region on a web page and send it to the Agent, use the built-in browser's annotation feature instead.

Follow up and adjust direction

While a task is running, you can send follow-up input to adjust the direction, such as narrowing the scope, adding constraints, asking for an explanation first, stopping the current action, or having the Agent continue with new requirements.

New input can wait in the queue. To move the current input to the front, press Command + S on macOS or Ctrl + S on Windows so the Agent handles it first.

Asking the Agent to stop the current action via a message counts as follow-up input. Pausing, resuming, and cancelling Goal mode, and stopping background tasks, are handled in their respective task controls—see Task Execution and Review.

Common commands

Type / to invoke built-in commands or saved extension commands and reuse fixed workflows. Common commands include:

Command	Purpose	Details
/compact	Compact the current context; view the compaction info in the right panel afterward	Right panel
/new	Start a new session	Project and Session Management
/fork	Fork the current session, deriving an independent copy with the full conversation history
/export	Export the session
/plan	Enter plan mode: the Agent proposes an execution plan first and only makes changes once confirmed	Task modes
/goal	Start goal mode and set an ongoing objective
/swarm	Use multiple subagents to work on a task in parallel
/undo	Undo the most recent message without reverting code changes; opens the selector without an argument	—
/status	View the current session status	—