AI agent: every page, round 2
From his critique of round 1 (994b45df). The switches sit in the job pages, as picked. Every page of the model is drawn at full size, at 1440 and on a phone, in the real shell. Pick a screen below; each has its own link. Nothing is built until the next critique.
His five points, and where each shows
| Point | Where it shows |
|---|---|
| 1. No coloured text. Text is ink or muted ink; state is a dot, an icon or a control. | Every screen. State words sit in ink beside a coloured dot (Profile's grid, Conversations' channels). Actions are shaped buttons or underlined ink, never tinted words. Warnings are ink on a pale ground with an amber icon (Comments). Remove sits in a menu. |
| 2. No group headers in the area panel. | The AI agent panel on every screen reads Profile to History as one list; a hairline marks where the jobs start and end. The same on every area in Every area's panel. |
| 3. An icon on every panel item, in every area. | The AI agent panel on every screen; all seven areas with panels in Every area's panel. Built app-wide after his look. |
| 4. Every proposed screen, desktop and phone, one switcher, each linkable. | The ten screens below: Profile, Topics, Knowledge, Actions, Guardrails, Conversations, Comments, Moderation, Playground, History. |
| 5. No fixed printed line standing in for a behaviour. Instructions the operator writes; the AI writes the words; a fixed fallback only when the AI cannot run, said as such. | Comments: the private message's close becomes "How the message closes" (instructions), the reply word shows the flow that listens for it, the fixed text is "Used only when the AI cannot write the message", with its count. The same in Topics (the hand-over), Conversations (asking for a name, a figure with no source, the public reply when the private message fails). |