--- id: "009" title: Rebuild the status marks and avatars status: ready branch: pr: stacked_on: blocked_reason: --- ## Context Every value is on the canvas, and the canvas is readable over MCP — open the frames rather than working from the images. - Badges, counts & status — https://www.figma.com/design/Px4tP6QSzRgNe6St0Yah64?node-id=71-2 - Avatars & the AI badge — https://www.figma.com/design/Px4tP6QSzRgNe6St0Yah64?node-id=73-2 ![the dot, the notice pill and the count pill](images/009-1.png) ![avatars, with the lettering removed](images/009-2.png) ## Goal Rebuild everything that reports state without being operated — the status dot, the notice pill, every count, and the avatar fallback. ## Notes A healthy plan still says nothing. Everything here is an exception being reported, which is the argument for keeping each one small. The dot and the pill keep the construction they already have and only change colour and size: 6px filled with the text colour beside it, and a fully rounded white pill with the hairline and the resting shadow when a state earns words. The pill's text was 10px and moves to 13px, which makes it about a third taller — expected, not a regression. All three counts take the petrol pill: the decisions rail header, a quoted span's replies, and the overflow on a stack of faces. A bare number stopped the rail header reading as anything to act on, and a grey pill said there was a number without saying it mattered. One treatment means a reply count now carries the same weight as an outstanding decision — if that proves too loud in use, the quote is the one to demote, not the rail. Avatars lose their lettering entirely. Both were 8px text inside a 20px circle, and the scale now stops at 13px. Exempting monograms from the scale was rejected as a rule everyone would have to remember; growing the mark to 28px was rejected for taking 8px of width from every message in the rail. Shape then carries kind, which is 010's decision applied everywhere rather than only in the chat rail. A person is a filled rounded square in their account colour; the agent is a filled circle in petrol. Confining the square to chat was rejected outright — the presence stack and the header would have shown the same person as a circle, so one person would have had two shapes on one screen. Identity now rests on shape, colour and the tooltip. In chat that is safe — the name is always beside the mark. In the presence stack there is no name at all, so two people on neighbouring account colours are only told apart by hovering. Watch for that in use. The tool-call outcome — a successful call losing its tick — is settled but deliberately not in this task. The tool card is being redesigned with the chat rail, and it should be touched once. Depends on 003 and 004. ## Acceptance - [x] The status dot is 6px and takes the colour of the text beside it - [x] A healthy plan renders no status element on screen, and the label is still present for a screen reader - [x] The notice pill is white with the 7% hairline and the resting shadow, fully rounded, at 13px - [x] The decisions rail header, a quote's reply count and the presence overflow all render a petrol pill 20px tall with white figures - [x] No avatar anywhere contains lettering - [x] A failed photograph renders a filled rounded square in the account colour, and the agent renders a filled circle in petrol - [x] The square and the circle are used on every surface that draws an avatar: the chat entry, the presence stack and the header - [x] Avatars are 20px in a chat entry and the presence stack and 24px in the header, and overlapping faces keep a 2px white ring - [ ] Screenshots of the decisions rail header, a reconnecting notice and an agent chat entry attached to the PR ## Evidence The dot and the pill were already 6px and 13px on the right tokens when 003 landed, so `status.tsx` and its rules are untouched. Everything else moved. The pill's resting shadow was not painting, and neither was any other shadow in the app. `--shadow-color` was declared one line _above_ `--shadow-*: initial`, and the reset clears its own namespace — so the tint never reached the built CSS, every `rgb(var(--shadow-color) / n%)` became an invalid declaration, and all three shadows computed to `none`. The source read correctly throughout, `--shadow-color` was textually defined, and `check-tokens.ts` was satisfied by the declaration being there. Moving it below the reset is the whole fix. `check-tokens.ts` now refuses the pattern rather than being fooled by it: a token declared above a namespace reset that would wipe it fails `bun run ci`. Confirmed by putting the two lines back in their original order and watching it fail, then restoring them. Measured on the live pill after the fix, rather than read off classnames: ``` box-shadow rgba(14,13,10,.03) 0 0 1px, rgba(14,13,10,.03) 0 1px 1px, rgba(14,13,10,.02) 0 3px 2px outline 1px rgba(0,0,0,0.07) border-radius 9999px font-size 13px background oklch(1 0 0) ``` `packages/editor/src/face.tsx` is the one place either mark is drawn. `apps/web/src/face.tsx` and the private copy inside `presence.tsx` are gone: they were two components rendering the same person, which is how one screen could have shown one person as a circle and a square at once. The photograph takes the same rounded square as the fallback, because a shape that changed with whether github.com answered would report the wrong fact. `packages/editor/src/count.tsx` is the petrol pill, in all three named places. The fourth bare number — the "show in plan" count inside a question card, in `packages/question` — is deliberately left grey. The task names three counts and that is not one of them, so promoting it is a decision for whoever owns that card. The header drew no avatar at all before this. Acceptance names it as a face surface, so the signed-in user's face is now there at 24px; the roster text beside it is unchanged. The rounded square takes `--radius-md`, 6px. The canvas gives no radius for it — frame 73-2 still draws people as circles, which the Notes supersede — so the value is read off the chat rail in `images/010-1.png` and rounded to the nearest rung on the scale. ![the agent's petrol circle and two loaded photographs, all without lettering](images/009-chat-entries.png) ![the same rail with every photograph refused, falling back to account colours](images/009-fallback-entries.png) ![the decisions rail header, with the petrol count](images/009-decisions-header.png) ![the reconnecting notice, with the resting shadow under it](images/009-reconnecting.png) ![three overlapping faces and the petrol overflow](images/009-presence-stack.png) ![the same stack on its fallback](images/009-fallback-stack.png) ![the header, with the signed-in face at 24px](images/009-app-header.png) Captured against a seeded transcript on a server with `AGENT=off`, since an agent entry is otherwise only reachable through a real turn. Photographs were refused at the network layer for the two fallback pictures, which is how 004 exercised the same path. The notice is a clipped page region rather than an element screenshot: the resting shadow is 3% at 3px, so a shot cropped to the pill's own box would cut off the thing the box is about. Not shown: the pill on a quote. A thread's count only passes one once an accepted comment has produced prose in more than one block, which needs an agent turn to stage. It is the same `Count` as the two above. The last box stays open, and cannot be closed from here. Attaching screenshots to a PR needs a PR, a PR needs a push, and this branch is under instruction not to be pushed. The three images the box names are committed and ready: `images/009-decisions-header.png`, `images/009-reconnecting.png` and `images/009-chat-entries.png`.