v0.12.52: @newsroom.writer: Commander's intent is a behaviour policy

A new article written here from a voice memo, on Josh Devon's "Your Agents Need Commander's Intent" (Sondera): agents do not go rogue, they fill unstated gaps, so limits must be written about effects, compiled into checkable rules and enforced outside the agent. Twelve of Sondera's terms joined to this site's vocabulary as a graph; where this site goes further (the business layer: function, process, relationship, organisation) and where Sondera does (plain English compiled to Cedar and proved). Five rules written in Cedar (Sondera's published example, this session's rules 9 and 7, a row of the skill behaviour policy, two business-layer rules) and fourteen requests evaluated with cedarpy 4.12.2, all as intended. The article's sources checked against Anthropic's, the UK AI Security Institute's and the Army's primary documents; two items unconfirmed. Two figures.

Also a one-line fix in the wire's styles, for @newsroom.wire: the Industry Wire's edition of 14 October made its page scroll sideways on a phone (596 px at a 390 px width), because one story card's minimum content width stretched the single-column stories grid. .stories now uses grid-template-columns: minmax(0, 1fr) and .story gets min-width: 0 and overflow-wrap: anywhere. The release gate's wire test caught it on live vault data; it passes again, as do the graphs and themes tests.