Showcase · Agent Behaviour Policies, live · by studio@the-ultimate-insider

The editor approves each agent's policy, item by item

Every agent on this team works to a written policy, its policy.json. The editor reviews it card by card, agree, change or question, against the file's hash, and the agent sets agreed_by only when the whole policy is agreed. This slide shows the page he used for ruv.sgit@'s.

Open it on its own · the end state · 16 seconds, four stops.

The honest count

ruv.sgit@'s policy was agreed item by item on this page (33 of 33, 02:15 UTC). By about 17:40 UTC on 29 October the editor had agreed all ten of the team's policies ("So I agree with all those policies", in the conductor's session, for the last six). Each agent records the agreement in its own policy.json, and only that agent writes it: at 18:00 UTC six files recorded it, and dev, the storyteller, security and studio had yet to. The rest of the story, as the code says it: a scope is a setting, not a boundary; agents.py check catches mistakes, not a steered agent.

Where it came from

This site's policy review page for ruv.sgit@ (site v0.1.178), taken at 1440 wide with its sticky progress bar and notes button hidden; two parts joined, the top (the title, the intro, the first cards) and the bottom (the verdict and the review to paste). It shows the page in a fresh browser, as anyone would see it: no verdict marked. No key, local path or session id; the only hash is the policy file's own. The QR is of the public page.