[currentbits]

The app gets a visual pass so it is calmer and easier to read, in both light and dark. Every change was checked against real screenshots of the running app, reviewed each round by a three-model design council.

What changes for the user

• One reading column and one title style on every page. Memory, Routines, Activity and Needs you gain the serif page title, and each page's main action sits on its title line (Add memory, Teach a skill, New routine). Pages with their own title no longer repeat it in the top bar.
• Easier on the eyes: 14px body text and 32px controls, brighter quiet text in dark mode, firmer dark card edges, and far less monospace. One blue accent for primary actions, toggles, focus and selection; amber is kept for things waiting on you.
• Lists that scan: Memory and Skills are grouped lists with dividers. Every row uses one title style, and meta lines are quieter (Memory drops the default "added by you"). Text actions ("Check", "See the steps") have a soft underline and come last in the line.
• Routines: the task is the row title and the schedule sits on the meta line. Each row shows a toggle, Run now and a ⋯ menu (Rehearse, Edit, Delete). On phones the actions move under the text.
• Settings: notes under each card become a muted footer row of that card. Selects use line chevrons, and ⌘F shows as a key chip in the search field.
• Bot page: the save bar sticks only while there are unsaved changes. The subtitle shows the bot's role, the AI account tiles are flat and one line each, and Test sits beside the Model field.
• Chat: a chat opened from a bot keeps that bot in the header and sidebar across the section tabs. The search collapses to an icon (⌘F opens it), the private-chat toggle is a lock, and the summary is a quiet strip.
• Activity gets column headings and drops the Tools column when no run used a tool.
• Fixes along the way: context menus render on the page body (a view's entrance animation used to pin them to that view), link underlines that never rendered (an old text-decoration: none won), and code chips that no longer split a command.

Tested: tsc -b passes, and scripts/ui-flows.mjs (updated for the routine ⋯ menu) passes all 42 checks against a fresh server, with no page errors. Reviewed from 1280×820 light and dark screenshots of the web UI. Not checked visually: onboarding, approval cards, the quick-ask panel, and the native macOS/Windows/Linux shells. The base size and control height change app-wide.

Found on the way, not fixed here: numeric cron weekdays (1-5) run Sunday to Thursday, because the cron crate counts 1 = Sunday. Suggested as separate work.

Council review

Rounds: 32. Stopped at the user's call: from round 15 the scores held at 7.5–8 while the reviewers' requests began to undo each other.

Delegate	R1	R2	R3	R4	R5	R6	R7	R8	R9	R10	R11	R12	R13	R14	R15	R16	R17	R18	R19	R20	R21	R22	R23	R24	R25	R26	R27	R28	R29	R30	R31	R32
Claude Opus	6/10	6/10	7/10	7/10	7/10	7/10	7.5/10	7/10	7/10	7/10	7/10	7/10	7/10	7.5/10	8/10	7.5/10	8/10	7.5/10	7.5/10	7.5/10	8/10	8/10	7.5/10	8/10	8/10	7.5/10	7.5/10	7.5/10	8/10	8/10	8/10	7.5/10
GPT Sol	7.2/10	no score	7.3/10	7.5/10	7.5/10	7.5/10	7.7/10	7.8/10	7.5/10	7.5/10	7.7/10	7.8/10	7.7/10	7.7/10	8/10	8/10	7.8/10	8/10	7.6/10	7.7/10	7.8/10	8/10	7.8/10	8/10	8/10	8/10	7.8/10	7.5/10	8/10	7.8/10	8/10	7.6/10
Gemini Pro	7/10	7/10	7/10	7.5/10	7.5/10	6.5/10	6.5/10	7.5/10	7.5/10	7.5/10	7/10	7/10	8/10	7/10	7.5/10	6.5/10	7.5/10	7.5/10	no score	no score	no score	no score	no score	no score	no score	no score	no score	no score	no score	no score	no score	no score

Gemini was unavailable from round 19 (its API prepaid credits ran out).

The demo account switched from fake-AI test text to realistic content from round 12.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-05T19:29:01.069883Z	8101ece	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8101ece8a3

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".