Skip to main content
Slash commands are short instructions that change how an agent behaves. You send them exactly like a normal message — they go in the message field of a stream request. The agent runtime sees the leading /, parses the command, and dispatches it instead of treating the text as a prompt.

How they work

Every slash command is a string you pass to sdk.agent.stream():
The runtime reads the command name and any arguments that follow it. Most commands also accept subcommands (for example /plan run or /goal pause).

What you get back

Some commands return action events in the stream — structured events like task.completed or action that report state changes as they happen. Others return a plain inline text response, similar to a normal chat reply. The table below notes which is which. All eight commands reuse the same threadId you already use for conversations, so command state and message history stay connected.

All eight commands

Next steps