agentleFS
fluxnative-appfluxnative-app/fluxnative-claude/skills/preview-verification/SKILL.md

preview-verification

Skill0 starsChanged 10 days agoMITView on GitHub

Use after changing a FluxNative app or landing page, or when the user asks how a FluxNative screen looks or behaves, why it is blank or why a button does nothing: build the preview, then look, tap and read logs in the live preview, and report only what that evidence shows.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Not a substitute for run_checks: compile errors, lint and tests come from there.

    line 12
  2. Very specific:

    Every tool here takes the project's id as project, the same id on every call.

    line 14
  3. Distinctive:

    Never report a screen as working because the build compiled or a write succeeded.

    line 49
  4. Distinctive:

    Never claim native behavior (purchases, push, camera, sensors) from the web preview.

    line 50
  5. Distinctive:

    Never type real personal data, real passwords or real payment details into the preview.

    line 51

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Check a FluxNative project in the live preview

When to use

  • You changed a screen, a flow or styling and need to see the result.
  • The user reports something visible: a blank screen, a broken layout, a control that does nothing, wrong data.
  • Not a substitute for run_checks: compile errors, lint and tests come from there.

Every tool here takes the project's id as project, the same id on every call; find it with list_apps (the app-workflow skill).

What the preview is

FluxNative runs the project's committed source in a real headless browser. An app runs as its Expo web build inside a phone or tablet frame; a landing page runs as the built site. The preview shows layout, navigation, copy, state and network calls. It does not run native modules (camera, push, in-app purchases, biometrics, native maps). For those, say what the preview could not show; a web fallback is not proof.

The preview runs the real app: a tap or a submitted form can reach the app's connected backend and change its data. Use test input the user gave you.

Procedure

  1. Build. build_preview compiles the current source and returns diagnostics. Fix compile errors before looking at anything.
  2. Reach the screen the way a user would. preview_view with route (for example /login) opens a route directly. For a flow, start from its entry screen and tap through it with preview_act. A screen that only works when opened by URL is not reachable.
  3. Read the structure, then look.
    • preview_view with mode elements (the default) lists the interactive elements with ref ids. It leaves out headings and plain text.
    • preview_view with mode text returns the visible text: use it for copy, empty states and error messages.
    • preview_view with find locates an element by its label or role and returns its ref.
    • preview_view with mode screenshot returns an image, which costs vision tokens. Take one per state you need to judge visually, not one per step; region zooms into one part. Hosts that render MCP Apps show the preview inline.
    • preview_view with inspect (a name, ref or coordinate) says where an element is written and how it is styled; sources adds file and line to the tree. Use them to go from screen to code.
  4. Act. preview_act with action tap, double_tap, long_press, type, scroll, swipe, drag, key or fill. Target by name (the visible label, preferred), then ref, then coordinate; a swipe takes a target and a direction, a drag two points. preview_act with action navigate and a route moves within the app. Read again after every action that should change the screen.
  5. Set the state a screen needs. preview_act with action reset_storage clears the preview's stored data (a first open); seed_storage with entries (each a key and a string value) stores app data, never a session; set_clock with at, advance or timezone moves the clock for a reminder, a streak or an expiry. The app restarts on its route, except after a clock advance.
  6. Vary the device.
    • An app: preview_view with device iphone-15-pro (the default), ipad-mini, pixel-8, iphone-duo (the foldable shut, a short screen) or iphone-duo-open (open, near-square); or preview_act with action resize and preset mobile, tablet, android, foldable or foldable-open, orientation landscape, colorScheme dark. Check dark mode and a small phone for any screen you restyled; the open foldable ignores orientation locks, so check it in landscape too.
    • A landing page: preview_view with device desktop (the default), ipad-mini and iphone-15-pro; or preview_act with action resize and preset desktop, tablet and mobile. Nothing may scroll sideways at any width, and every link must land somewhere real.
  7. When something is wrong, read the logs before the code. preview_logs with kind console and errors_only shows runtime errors. preview_logs with kind network lists requests; pass a request_id to get one call's status, headers and body.
  8. Fix and repeat from step 1 until the screen and its states match the intent.

States worth checking

  • Loading, empty and error states, not only the happy path with data.
  • Long names and large numbers, so text wraps instead of clipping.
  • Signed out and signed in, when the app has sign-in.
  • Back navigation from a pushed screen, and switching tabs and coming back.

Never

  • Never report a screen as working because the build compiled or a write succeeded.
  • Never claim native behavior (purchases, push, camera, sensors) from the web preview.
  • Never type real personal data, real passwords or real payment details into the preview.

Report

  • What you checked, on which device and color scheme, and what you saw.
  • Errors found in the console or network logs, and what you changed because of them.
  • What the preview cannot verify and still needs a real device.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.