agentleFS
imbue-aiimbue-ai/sculptor/.claude/skills/auto-qa-iphone/SKILL.md

auto-qa-iphone

Skill246 starsChanged 5 months agoMITView on GitHub

QA the Sculptor mobile web UI on a real iOS Simulator, driven headlessly from a Mac. A single CLI boots a notched iPhone, launches the local frontend server (parsing its port — no hardcoding), opens the app in MobileSafari, taps/swipes via idb, and screenshots each step. Includes the Add-to-Home-Screen standalone flow — the only way to verify notch / status-bar / home-indicator safe-area rendering (env(safe-area-inset-*) is 0 everywhere except real iOS). Use when visually verifying mobile/responsive changes on an iPhone.

  • Installs packages

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

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

    If nothing is found, fall back to launching --command (default just frontend-custom), parse its port from the log, and verify it actually …

    line 85
  2. Very specific:

    Every action prints a screenshot: path and an <img> tag — paste the tag into chat to show the user (see "Context management").

    line 99
  3. Distinctive:

    Verify each screenshot matches the expected step before trusting the next.

    line 131
  4. Distinctive:

    If the home screen is still showing afterward, the tap missed the artwork — re-locate and tap again.

    line 142
  5. Very specific:

    iOS caches the launch config at add-time: After changing index.html's <head> (meta tags, theme-color, icon), remove and re-add the clip — …

    line 240

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

iPhone Simulator QA (Simulator + idb)

This is the iPhone sibling of /auto-qa-changes. Instead of a headless Chromium at desktop size, it drives a real iOS Simulator so you can see how the mobile web UI actually renders on a notched iPhone — including the standalone PWA chrome (the app sets apple-mobile-web-app-capable, viewport-fit=cover, and a black-translucent status bar in sculptor/frontend/index.html).

Why a real simulator and not Playwright: env(safe-area-inset-*) — the insets the mobile shell uses for the notch/Dynamic Island, status bar, and home indicator — are 0 everywhere except a real iOS surface. Desktop browsers, Playwright, and the /auto-qa-changes Chromium harness all report 0. The only faithful check is Add-to-Home-Screen + a standalone launch on a notched device.

Everything is driven through one CLI:

.claude/skills/auto-qa-iphone/scripts/iphone_sim.py

Stdlib-only Python (no uv/venv needed to run it); it shells out to xcrun simctl, idb, and just.

Prerequisites (one-time, slow)

  • macOS + Xcode with an iOS Simulator runtime. setup checks for one and, if missing, prints the (~9 GB) download command xcodebuild -downloadPlatform iOS — it will not download automatically unless you pass --download-runtime.
  • Homebrew + idb (drives taps/swipes the simulator CLI can't). setup creates an idb venv at ~/.cache/sculptor-iphone-qa/idb-venv and brew install idb-companion for you.
  • A working frontend dev env (just rebuild once if this is a fresh checkout), since the default server command is just frontend-custom.

Quick start

Set a short alias and a screenshots dir (defaults to the workspace attachments/, so images render in chat and survive for MR reuse):

SIM=".claude/skills/auto-qa-iphone/scripts/iphone_sim.py"

Step 1: Boot the device (idempotent)

python3 "$SIM" setup            # default device: iphone-16-pro (notched)
# Other presets: --device iphone-16-pro-max | iphone-se (non-notched control)

This creates/reuses a SculptorQA-<preset> simulator, boots it, connects idb, and prints the UDID. Re-running is safe (it reuses the device). To watch it live: open -a Simulator.

Step 2: Open the app

Have your dev server running first (e.g. just frontend-custom or just start in your own terminal — it works reliably there). Then:

python3 "$SIM" open

How open/serve resolve the URL (no port is ever hardcoded — the dev port is hashed per checkout):

  1. --url URL / --port N — attach to exactly that server.
  2. Otherwise auto-detect: probe listening ports for a running Sculptor SPA and attach to it (a vite dev server is preferred over a packaged/production app build; it warns when it could only find the latter). This is the normal, reliable path.
  3. If nothing is found, fall back to launching --command (default just frontend-custom), parse its port from the log, and verify it actually serves before using it.

The simulator shares the host network, so http://127.0.0.1:<port> is reachable.

Launching from inside a Sculptor agent is flaky. just frontend-custom drives Electron, which (a) races to load its renderer before vite is ready and dies under --unhandled-rejections=strict, (b) refuses a second instance via the .dev_sculptor single-instance lock, and (c) inherits a leaked SCULPTOR_API_PORT from the surrounding session. Prefer keeping your own dev server running and letting open auto-detect it (or pass --port/--url). Use --launch to force the fallback launch anyway.

Every action prints a screenshot: path and an <img> tag — paste the tag into chat to show the user (see "Context management").

Step 3: Drive the UI and narrate (see protocol below)

python3 "$SIM" screenshot --label home          # capture current screen
python3 "$SIM" tap --frac 0.5 0.93              # tap by screen fraction (0..1)
python3 "$SIM" swipe 201 620 201 320            # swipe by points
python3 "$SIM" describe                          # a11y tree of the foreground app

Step 4: Verify safe areas via Add-to-Home-Screen (the important part)

Safari always shows its own chrome, so the standalone status bar and true safe-area insets only appear after adding to the home screen and launching.

First, make sure Safari's bottom toolbar is visible: if the page auto-focused an input, the on-screen keyboard hides the toolbar (and the Share button). Dismiss it by tapping an empty area, then screenshot to confirm the toolbar is back:

python3 "$SIM" tap --frac 0.5 0.42              # tap empty space to blur the input

Then drive the flow (it screenshots after every tap because SpringBoard and the share sheet aren't in the accessibility tree, so they can't be located):

python3 "$SIM" add-to-home-screen               # Share -> swipe -> Add to Home Screen -> Add

Verify each screenshot matches the expected step before trusting the next. Then launch the clip. The icon's position is not fixed, so locate it first and tap its artwork center (the text label sits ~40 pt below the artwork — aim at the glyph, not the label) via a describe-screenshot subagent, then:

python3 "$SIM" screenshot --label home          # capture the home screen
# (locate the Sculptor icon ARTWORK center as a fraction, then:)
python3 "$SIM" launch-icon --frac 0.39 0.26 --settle 6

If the home screen is still showing afterward, the tap missed the artwork — re-locate and tap again.

Step 5: Clean up

python3 "$SIM" teardown        # stops the managed server + shuts the sim down
# python3 "$SIM" teardown --delete   # also delete the simulator device

Available commands

CommandWhat it does
setup [--device P] [--download-runtime]Ensure runtime + idb; create/boot SculptorQA-<P>; connect idb; print UDID.
detectList running Sculptor servers (tagged dev/app) and show which one open would attach to.
serve [--port N | --url U | --command C | --launch]Resolve a server URL: attach to --port/--url, else auto-detect a running one, else launch --command.
open [--port N | --url U | --command C | --launch]Same URL resolution as serve, then simctl openurl → screenshot.
screenshot [--label NAME]Capture the current screen to a numbered PNG.
tap X Y [--frac]idb ui tap. --frac treats X/Y as 0..1 fractions of the screen. Auto-screenshots.
swipe X1 Y1 X2 Y2 [--frac] [--duration S]idb ui swipe. Auto-screenshots.
describeidb ui describe-all — JSON a11y tree (foreground app only).
add-to-home-screenDrive the AHS flow with a screenshot after each tap.
launch-icon [X Y] [--frac]Tap the home-screen clip to launch standalone; screenshot.
remove-home-screenGuidance for removing the clip (iOS caches launch config at add-time).
teardown [--delete]Stop the managed server; shut down (and optionally delete) the device.
statusPrint the device/server/idb state (JSON).

Shared options (after any subcommand): --screenshots-dir PATH, --settle SECONDS. State (UDID, point size, server pid/url, screenshot counter) persists to <screenshots-dir>/.iphone-sim-state.json, so commands chain without re-passing the UDID.

Context management: never Read screenshots directly

Simulator screenshots are native-scale PNGs (iPhone 16 Pro is 3× → ~1206×2622), each large. Never call the Read tool on a screenshot yourself — a few will blow out your context. Same three rules as /auto-qa-changes:

  1. Display to the user via the <img> tag the CLI prints. This is free for your context — it renders on the user's side; no image data enters yours.
  2. Verify state programmatically when you can. describe (a11y tree), status, and the success/failure of each command tell you a lot without pixel inspection.
  3. When you genuinely need visual inspection (alignment, what text appeared, whether the status bar overlaps content), delegate to a subagent with a narrow question:

    Agent(description="Describe screenshot",
          prompt="Read /abs/path/0007_standalone.png. Focus on the top of the
                  screen: does any header content sit UNDER the status bar / notch,
                  or is there correct safe-area padding above it? 2-3 sentences.")

    The image stays in the subagent's context and is discarded when it returns.

Narrated visual walkthrough protocol

Same as /auto-qa-changes. Every screenshot you take, you MUST:

  1. Display it with the <img> tag (absolute src, descriptive alt).
  2. Describe what you see — inferred from describe/status, or via a describe-screenshot subagent. Do not Read the PNG yourself.
  3. Call out issues — note the safe areas especially: content under the notch, a chat input hidden behind the home indicator, clipped headers.
  4. Announce the next action.

Format each step:

<img src="/abs/path/attachments/iphone-screenshots/0002_open.png" alt="Mobile workspace shell, Safari">

**Step 2: Mobile shell in Safari**
Single-column layout: header, chat stream, floating input, agent pager at the
bottom. Safari's own toolbar is visible (expected — standalone chrome only shows
after Add-to-Home-Screen). No obvious layout issues.

**Next:** add to home screen and launch standalone to check the real status bar.

Driving taps without an accessibility tree

  • Coordinates are points, not pixels. iPhone 16 Pro is 402×874 pt; screenshots come out at 3× (1206×2622 px). idb wants points.
  • Prefer --frac. Spot the target in a screenshot, express it as a fraction of the image (e.g. a button 90% down the screen → --frac 0.5 0.9), and the CLI multiplies by the device's point size. This is robust across devices.
  • WebView elements (inside MobileSafari) appear in describe; SpringBoard, the share sheet, and the home screen do not — drive those by coordinate and verify with the screenshot after every tap.

Gotchas

  • The standalone status bar only appears via the home-screen launch, never in Safari (Safari always shows its own chrome). You MUST go through add-to-home-screen + launch-icon to exercise apple-mobile-web-app-status-bar-style and the real safe-area insets.
  • iOS caches the launch config at add-time. After changing index.html's <head> (meta tags, theme-color, icon), remove and re-add the clip — reloading isn't enough. See remove-home-screen.
  • AHS coordinates are empirical and only ship for iPhone 16 Pro. They can drift across iOS versions; the flow screenshots between steps so you can catch drift and re-derive with tap --frac. For other presets, drive AHS manually.
  • env(safe-area-inset-*) is 0 on non-notched devices — --device iphone-se is a deliberate "insets ≈ 0" control, not a notch test.
  • Hot reload works in Safari, since just frontend-custom runs the vite dev server: edit a .tsx/.scss and re-screenshot to see it. But a standalone (home-screen) clip does not pick up <head> changes without a remove/re-add.
  • The simulator shares the host network, so http://127.0.0.1:<port> works (unlike a physical device). No tunneling needed.
  • The server runs detached (own process group) and survives across turns; teardown stops it. status shows whether it's alive.

Cleanup

teardown stops the managed server and shuts the simulator down (keeping the device for next time; --delete removes it).

NEVER delete screenshot files — they're referenced by <img> tags in the user's chat history and may be attached to MRs. Leave them in the screenshots dir.

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.