Skip to content

AI Compass / Quick reads

"Done" is a command with an expected output

Automation Prompting

Anyone who hands work to an AI should write down beforehand how they will recognise "done", and not let the AI itself provide the proof.

In depth: "Done" only with evidence – a check command instead of self-reporting

An agent quickly reports “done”. The open-source skill Unlazy turns that around: before the work, a gates file lists for each criterion the command that proves it, the expected output and an evidence line. A tick without evidence counts as open (AI LABS 20 August). After the work, the main agent runs the commands itself, not the agent that built it. For large projects, the skill Wayfinder clarifies the open decisions beforehand. The result is then checked against this answer key (14 August). How little a green tick says on its own is shown by another video from the same channel: all 11 checks were green, yet the order form had not been tested at all (2 October).

Tool and alternative

Unlazy and Wayfinder are free (third parties, read them first). According to the README, the Unlazy gates template can also be copied without installing anything. A simple list in the task does the same job.

Sources

Advertising in the source: DataForSEO and SerpApi (20 August), Upstash Box (2 October), each plus a paid community.

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