mirror-screen
Skill5 starsChanged 13 days agoApache-2.0View on GitHub
Mirror one screen of the connected app in HeroCanvas (app/) from the app's code - its components with the code's names and props, its data as mock data, its texts in every language - and check it against the code. Use when the person asks to build, mirror, add or fix a screen or a component of their app.
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- connect-appSkill
- new-personaSkill
- new-proposalSkill
- update-herocanvasSkill
The file
Files
Mirror a screen
A screen in HeroCanvas is a mirror: the app's screen rebuilt in React, close enough to the code that a developer reading Inspect sees their own components, props and tokens. It's what designs start from (proposals change it), so it must be what the app is today, not what it should be.
1. Read the screen in the app's code
The screen's file and every component it builds, down to the app's design system. Note: the components and their parameters, the tokens they use, the texts (their keys), the data they show and where it comes from, the states (loading, empty, error, a long text), what each tap does. Read, never edit the app's code.
2. Components first
For each component the screen uses, reuse the mirror's if it exists (app/components). The platform's own views
(SwiftUI's List and Section, NavigationLink, Picker, Toggle…) come from herocanvas/ios, not rewritten. If not, write
it there:
- Its name and props are the code's (
onPressed, notonClick, if the code says so). A prop the code doesn't have (something only the simulator needs) is declared withdeclareProps(src/design/simulatorProps.ts):simulator(only here),differs(a design option the code lacks),look(it looks unlike the code). - Pure: props in, callbacks out; no navigation or app state inside.
- Instrumented:
useComponent(name, props)on its root (fromherocanvas), so Inspect and Layers name it. - Styled with the app's tokens (
var(--app-…)), never raw values another component also uses. - A story beside it (
<Name>.stories.tsx,src/design/stories.ts): its states, for the gallery. - Every text in an element of its own, in one of the app's type styles (
font: var(--app-font-…)).
3. Data and texts
Mock data in one module under app/data/, derived from the persona (?persona=), never numbers written in the
screen that another screen also shows. Texts by their key in the app's translations (app/i18n.ts), in every
language the app has: npm run texts -- --from <its translations> converts them into one JSON per language
(app/i18n.<lang>.json), never retyped; images the screen needs, npm run assets -- --from <its catalog>.
What only the platform draws (a MapKit map, a Swift Charts chart, a video, a web view) is drawn here as close as
the web allows: a chart's marks with SVG, a map as its colors and a pin. Say in the component's comment what it
stands for. Icons are the app's: an SF Symbol or an app's own symbol as sf:<name> (npm run symbols makes any that's missing:
the console says which), else the closest Material Symbol.
4. The screen
In app/screens/, reading the layout from the simulator (useSim().layout, usePadL, usePadR), never a
device's name or size. A screen that changes by size class (SwiftUI's horizontalSizeClass: a List on iPhone, a
Table on iPad) reads layout.horizontalSizeClass; one by Android's window size classes, layout.vw. A bar fixed at the top
(a navigation bar, sticky while the page scrolls) spans the screen's whole width: take the side margins back with
negative margins and pad its content by them (the tour's Screen, src/tour/TourKit.tsx), or the page shows beside it
as it scrolls under. Sheets and dialogs through useSim().openSheet / openDialog / useActionSheet (they become
menus and dialogs on large screens by themselves). States the person can try go in the page's manifest entry as
controls (Scenarios). Then status: 'done' in app/pages.tsx.
5. Check it
npm run shot -- "/<route>", on a phone and a tablet, light and dark: look at the pictures beside the app (a
screenshot of the real app if the person has one). npm run typecheck, npm run lint, npm test. Show the person
the pictures and what differs from the code, if anything.
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.

