Agent RX
Agent RX works inside a terminal project directory: it can read and change files, run commands, and retain session records. You choose the model route; the task result is the resulting project state, not the confidence of the response.
Choose a starting point
- Open a workspace to confirm the project and inspect available controls before giving work.
- Give a bounded task to name a small change and its acceptance check.
- Use the command reference to distinguish shell entry points from commands inside RX.
Before you give it work
Save your own work and record existing changes. A request for read-only analysis is not an operating-system permission boundary. Ordinary sessions use the host workspace; do not assume a virtual machine or a universal sandbox. Files and command output read for a task can become model context. Review the selected route before sharing private material.
The distribution documentation describes provider routing and platform limits. Provider access and charges belong to that route and your account; an RX session is not a grant of model access.
Host workspace and bundled environments
The terminal workflow documented here uses the host project directory. It does not provision a bundled virtual machine, an embedded display or ephemeral stores. Do not assume that closing a session destroys files or isolates shell execution. Use a separately authorized environment when a task requires isolation, and inspect the active workspace before giving it access.
What completion means
Inspect changed files, review unexpected additions, and run the relevant project check. Session records and checkpoints help explain work; they do not establish correctness or guarantee file rollback. Keep the exact non-secret error if a task stops.
Read privacy and permissions for shared handling guidance and product handoffs when supplying reviewed source material.