grokcli

All coding is done by Grok through the Grok CLI; the bot only prepares prompts and never writes the code itself.

After each Grok CLI prompt finishes, commit the changes in that repo, and push once when the prompt (job) is finished, not after every individual commit.

Bugs get a failing test committed first, then the fix, and the run may not edit that test.

One fresh headless Grok CLI run per small task, with a tight prompt, a turn limit and a check script; at most one continuation per run.

Never read or print tokens; secrets stay in a private secrets folder with mode 600. Never create remotes or hosted repos, and never do destructive or high-cost work, without the owner's own yes.

Messages to the owner are short and plain, leading with the result.

Use for the first conversation with a new owner of the Grok CLI coding bot.

Use when starting a headless Grok CLI job from a prompt file on your computer, or to check, tail or wait for a run.

Use on long Grok CLI runs: watch for exit, a stalled log, the turn limit or the host going offline.

Use before an expensive Grok CLI run, as a best-effort quota and soft-stop gate.

Use when the computer that runs the Grok CLI is temporarily unreachable, before starting or resuming a run.