Run one task with rx ask

Use rx ask when you want to submit one task from your shell and receive its answer there. It is an agent invocation, not a read-only query or a shortcut for safely executing arbitrary text.

Before you start

Confirm your current directory with pwd, save existing work, and configure the intended provider separately. Check rx ask --help for installed options. The command can invoke a model, run tools, change project files, and use provider allowance.

Submit one task

Adapt this example only after confirming the repository actually contains parser tests and you are authorized to work on them. The example requests a proposal; that wording does not prevent writes or shell execution.

rx ask "Inspect the parser tests and propose a narrowly scoped correction. Preserve unrelated changes. Report the relevant check and its actual result."

The distribution documentation documents one-shot operation. Do not paste unreviewed issue text, web content, or a credential into an executable shell command. Shell quoting and agent instructions are different boundaries.

Check the result

Retain the command's exit status and its output. Read the proposed correction, inspect changed and untracked files, and check the actual project test result. An answer that names a test is not evidence that the test ran; compare the reported command and result with available execution output.

For a Git project, use git status --short and git diff, then inspect the affected files. Follow bounded-task review for the full review pattern.

If it stops

Preserve the non-secret error and current changes. Missing provider access is a setup problem; a failed project test is task evidence. Do not retry an uncertain external action merely because the shell returned. Check its real effect before another invocation. Use shared support if the result remains unclear.