sushiai-ui
Skill2 starsChanged 27 days agoMITView on GitHub
The sushiAI design code: tokens, type scale, the component catalogue and the recipes for new windows, taken from the shipped screens. Use before writing TSX or CSS for any change that adds or alters a screen, dialog, window, panel, menu, popover, list row, settings section or component in src/, and whenever Figma is unavailable. It replaces outside design skills for this repo.
In this repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- author-and-verify-extensionSkill
- autoreviewSkill
- cut-releaseSkill
- deslopSkill
- extension-manifest-changeSkill
- figma-to-codeSkill
- ship-prSkill
- sushiai-taskSkill
- sushiai-testingSkill
- ui-evidenceSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Every value here was read from the code.
- Distinctive: line 94
Describe the screen in words against the catalogue: before coding: which shell, which rows, which controls, the one primary action, the …
- Distinctive: line 97
Done when: every element in the description names a catalogue entry or a deliberate, stated extension of one.
- Distinctive: line 109
Done when: every difference is either fixed or written down as intended.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
sushiAI UI
The design code of the app as it ships today, so a new screen looks like it was always there. Figma is the source when the owner provides a concept; when there is none, this skill is the concept. It replaces every outside design skill (frontend-design, ui-ux-pro-max, design, ui-styling and the like) for sushiAI: do not load them, their taste is not this app's.
Every value here was read from the code. If the code and this file disagree, the code wins and this file is the one to fix.
The style in one paragraph
Dark, dense, quiet tool chrome. Near-black surfaces separated by one-shade
steps and hairline borders, not by shadows. One sage accent (--accent) for
the single primary action and "on" states; status is a 6px tone dot or a
16px tag, never a coloured card. Text is the system sans at 10-13px with
fixed pixel line-heights; only dialog titles go to 18-20px. Weights are 400,
500 and 600, nothing else. Uppercase 10px eyebrows and group labels carry
structure. Motion is almost absent. Plugins render inside the app's own
window and never take over the window frame or the left navigation.
It is not: a marketing page, a dashboard of cards, a gradient, a hero, a centred stack of big inputs, emoji, coloured panels, or copy that explains itself.
Rules
- Reuse, then extend, never fork. Find the nearest existing pattern in
references/components.mdand open its canonical file before writing a line of CSS. Use its classes. If it lacks a variant, add a modifier class next to it. If a pattern scoped to one area (pk-,np-,pd-) is needed in a second area, promote it to a sharedui-primitive insrc/ui/in the same change and move the old callers - do not copy it under a new prefix. - Spend tokens. Colours, radii, gaps and type come from
src/styles/tokens.css(references/tokens.md). CI rejects a raw hex insrc/styles/**/*.cssoutsidesrc/styles.cssandtokens.css. Never use avar(--x)that tokens.css does not define - an undefined variable fails silently. A colour the app does not have is not added in a feature change. - New CSS goes in
src/styles/components/<area>.css, imported once insrc/main.tsx, with a header comment naming the screen. That folder is what the CI colour check covers;src/styles.cssis legacy and only shrinks. - Scope inside dialogs. The legacy
.modal form,.modal label,.modal input,.modal h2rules restyle everything in a dialog. Write two-class selectors (.pd .pd-input,.np .np-head h2) so yours win; see the specificity section ofreferences/recipes.md. - Icons are lucide-react, sized by role (12 inline, 14 in a labelled
button, 15 in a nav rail, 17 in a picker row or the dialog close X).
Decorative icons get
aria-hidden. Agents use their own marks (Iconinsrc/PanelIcon.tsx), never an emoji. - Provenance and status sit at the name: a small icon or
Tagright after it, the full words intitle. No heading rows per category, no repeated status pills down a list. - Copy is short and states consequences. Sentence case, English, no
exclamation marks. A hint says what happens ("Tasks branch from here and
land back into it."), not what the control is.
·separates facts,…ends a label that opens a confirmation,→names a path through the UI. - Accessible by construction: real
button/input/select, roles on custom widgets (tablist,radiogroup,menu,menuitemradiowitharia-checked),aria-labelon icon-only buttons, a visible focus ring,role="alert"on errors. Tests and the evidence driver select by role.
Catalogue at a glance
| Need | Use | Canonical file |
| Dialog shell | DialogHost + .modal.<name> | src/dialogs/DialogHost.tsx |
| Two-pane settings window | rail + main (settings-* / pd-*) | src/app/SettingsDialog.tsx, src/ProjectSettingsDialog.tsx |
| Compact dialog (640px) | pk-head / np-head, footer | src/app/PanelPickerDialog.tsx, src/WorkspaceDialog.tsx |
| Page with title + scroll | ProjectPage (pd-title, pd-body) | src/ProjectPage.tsx |
| Labelled setting row | Field (pd-field) | src/ProjectGeneralTab.tsx |
| Choice list row | pk-row + pk-row-main + pk-text | src/app/PanelPickerDialog.tsx |
| Dropdown / account menu | picker-account-menu + pk-menu-* | src/app/PanelPickerDialog.tsx |
| Table of records | pd-table + pd-row (+ .head) | src/ProjectEnvironmentTab.tsx |
| Buttons | ui-button primary/secondary/ghost | src/ui/ui.css |
| Switch, tag, chip, banner | Toggle, Tag, Chip, Banner | src/ui/ |
| Status dot | ui-dot ui-tone-* | src/ui/ui.css |
| Key hint | kbd.pk-kbd | src/styles/components/picker.css |
| Error / empty | pd-alert, pd-empty, Empty | src/styles/components/project.css, src/app/Empty.tsx |
Full entries with sizes and snippets: references/components.md. Recipes for
a new settings section, dialog, project tab, menu and list row:
references/recipes.md.
Concept loop when there is no Figma
- Describe the screen in words against the catalogue before coding: which shell, which rows, which controls, the one primary action, the empty and error states, and the nearest existing screen it must sit beside. Done when: every element in the description names a catalogue entry or a deliberate, stated extension of one.
- Build it from those components, tokens only, CSS in its area file.
npm run build, then photograph it with$ui-evidence(driver inartifacts/, hidden window, throwaway profile). Capture the new screen and the nearest existing screen at the same window size.- Compare side by side: left edges and baselines line up, row heights
and gaps match the neighbour's rhythm, the same text role has the same
size and weight, one accent at most. Measure, don't eyeball: text ink via
Range, computed styles for anything that must match. Fidelity tolerances as in$sushiai-task: text width within 3%, shifts within 2px. Done when: every difference is either fixed or written down as intended. design-criticjudges the PNGs in fresh context. Fix and re-shoot; one full pass and one delta pass.
Green tests never prove a screen looks right. Open every PNG you cite.
Slop tells to refuse
Gradients or glows as decoration (the ones that exist carry meaning: the
pd-body scroll-fade mask, the hatched waiting stage, a loading shimmer, an
image's transparency checkerboard); emoji as icons; a 25px+ heading inside a tool pane (the legacy
.modal h2 is 25px - override it); a card inside a card; a centred column of
stacked full-width inputs where the app uses label-left rows; the same status
pill repeated on every row; a new colour, shade or shadow; mixed icon sizes in
one row; buttons of different heights side by side without reason; copy that
narrates ("Here you can manage..."), filler subtitles, title case
everywhere; a third copy of a menu, input or segmented control. More, with
the real counter-examples: references/recipes.md.
Before you hand it over
- Nearest existing pattern named and reused; no forked class family.
- No raw colour in new CSS; every
var(--x)exists in tokens.css. - New CSS in
src/styles/components/<area>.css, imported insrc/main.tsx. - Sizes come from the scale in
references/tokens.md; weights 400/500/600. - Icons lucide-react at the role size,
aria-hiddenwhen decorative. - Empty, loading and error states exist and use the catalogue classes.
- Keyboard: Escape closes the innermost layer only; focus ring visible.
- Nothing renders outside the app's own content area or over its nav.
- Built, photographed with
$ui-evidence, compared against its neighbour, judged bydesign-critic. ./node_modules/.bin/prettier --writeon touched files;npm run ci.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

