Agent RX command reference

Use shell entry points to start RX; use slash commands only after the workspace opens. This reference preserves the documented basic controls. Inspect /help for the surface and version you actually use.

From your shell

Command	Meaning and effect
rx	Open the terminal workspace; may initialize local session state.
rx ask "TASK"	Submit one task and print an answer; may invoke a model, modify files, and run tools. Replace TASK with your request.
rx --help	Read command-line usage without submitting a task.
rx --version	Read the installed binary version.
rx update	Install an update; changes the installed binary, not a diagnostic read.

Distribution documentation describes additional controls. Do not infer that every option there exists in an older installation.

Inside RX

The following example contains separate in-session commands. Enter them individually, not as a shell script and not as a single task.

/help
/models
/tools
/sessions
/checkpoint

Command	Scope and caution
/help	Available session commands.
/models	Model choices; provider discovery may use network access.
/tools	Tools exposed to the model, not a permission grant.
/sessions	Retained session/history controls; semantics can differ between terminal surfaces.
/checkpoint	Checkpoint controls; can create state. Not a read-only inspection or a universal restore command.

Tools and result checks

The documented default tool names are read, write, edit, glob, and bash. Inspect the installed inventory rather than assuming a fixed set. Shell activity and file writes use the configured execution environment; a restrictive prompt does not enforce host isolation.

Check the actual files and relevant project test after work. Retain errors and existing changes before retrying. For the first session, use Open a workspace; for review, use tool activity.