Skip to content

AI Compass / Quick reads

/goal: a second model decides when to stop

Automation Prompting

Letting an AI work unsupervised is justifiable when it is settled beforehand what "done" means and someone other than the AI checks it.

In depth: Let Claude work until the goal is met – with /goal

With /goal you give Claude Code a condition, such as “all tests in test/auth pass”. After each round, a small, fast model checks whether it is met. If not, Claude keeps working (docs checked). Important according to the docs: the checker only reads the transcript and runs nothing itself. So the condition must be something that Claude’s own output proves, with a check command and a limit such as “or stop after 20 rounds”. AI LABS builds a queue from this: one spec file per feature that is also the checklist, a queue.md with todo/building/done, one builder per branch and a checker with fresh context (18 August). The human reads and merges the pull request.

Tool and alternative

Included in Claude Code, without questions only in auto mode. Long runs use a lot of quota. In the video, the example ran for about 3 hours and was not yet finished; the channel gives no consumption figure. You do not need the GitHub, Supabase and Vercel from the video.

Sources

Advertising in the source: Sponsored by Hedra, plus a paid community.

Want to try this with your team? That’s what the AI Lab is for. See the AI Lab