agentleFS
simiancraftsimiancraft/simiancraft-skills/skills/playwright-camera-mask-testing/SKILL.md

playwright-camera-mask-testing

Skill8 starsChanged 5 months agoMITView on GitHub

Specialization of playwright-harness for CAMERA / segmentation testing in the browser: put a real PERSON in front of getUserMedia so MediaPipe / selfie segmentation produces an actual mask, then verify background-replacement / blur / shader effects by vision. Use when the task is "test the mask on web", "verify segmentation in the browser", "feed the camera a person fixture", "check a camera/background effect in Chrome", or any time a web app must SEE a human through the webcam without physical hardware. Read playwright-harness FIRST for script conventions and execution; this skill only adds the camera-feed pieces. The web analog of android-emulator-mask-testing. Validated on Linux/WSL, headless Chromium, against MediaPipe Selfie Segmentation.

  • Installs packages

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

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

    Keep dimensions even (yuv420p requirement) and resolution ≤ 720p.

    line 81
  2. Distinctive:

    Use the base's serving recipe (playwright-harness, "Serving an exported static site").

    line 117

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

Camera / Mask Testing: specialization of playwright-harness

Read playwright-harness first. The base owns prerequisites, the run pattern (write to /tmp/pw-*.mjs, run with playwright resolvable), and the drive/assert

  • WebGL-GPU patterns. This skill changes only how the camera is fed. MediaPipe selfie segmentation runs in WASM and works headless.

Prerequisites (in addition to the base)

ffmpeg, only to turn the still fixture into the looping .y4m Chrome plays as a fake camera; no system install needed, npm i ffmpeg-static in a /tmp dir gives a binary path.

This skill ships two fixtures (derivations and provenance in fixtures/README.md):

  • fixtures/person-web.png: a 640x480 webcam-shaped frame, full-body subject occupying ~82% of frame height, vertically centered, on a clean light-gray ground (#CDD0D4). Derived from android-emulator-mask-testing/fixtures/person-framed.png. Web presents the capture file 1:1 (no sensor crop), so do NOT reuse the Android fixture directly; its left-quarter framing compensates an emulator sensor shift that does not exist on web.
  • fixtures/office-empty.png: a 1280x720 empty office, the no-person negative case: a realistic scene (furniture, plants, windows, hard shadows) with nothing segmentable in it. Use it to mock the scene itself when the feed should contain no one.

The three camera modes (pick by what you are asserting)

  1. Green-ball mode (Chromium's built-in synthetic feed): proves the pipeline RUNS end to end (getUserMedia resolves, effects apply, no page errors). Useless for mask quality; the feed has no person.

    args: [
      '--use-fake-device-for-media-stream',
      '--use-fake-ui-for-media-stream',          // auto-grants the permission prompt
      '--autoplay-policy=no-user-gesture-required',
    ]
  2. Person mode (file-backed feed): proves mask SHAPE and compositing (person kept, background replaced/blurred). Add one flag; it requires --use-fake-device-for-media-stream to be present too:

    args: [
      '--use-fake-device-for-media-stream',
      '--use-file-for-fake-video-capture=/tmp/person-web.y4m',
      '--use-fake-ui-for-media-stream',
      '--autoplay-policy=no-user-gesture-required',
    ]

    The file must be .y4m (uncompressed YUV4MPEG2) or .mjpeg; Chrome loops it forever. Generate the y4m from the fixture at test time; it is too big to keep around (size = W x H x 1.5 bytes x frames; 640x480 x 2 s @ 15 fps ≈ 14 MB). No system ffmpeg needed: npm i ffmpeg-static in a /tmp dir.

    ffmpeg -loop 1 -i fixtures/person-web.png -t 2 -r 15 -pix_fmt yuv420p /tmp/person-web.y4m

    Keep dimensions even (yuv420p requirement) and resolution ≤ 720p. A malformed y4m (odd dimensions, wrong pix_fmt) makes Chrome play nothing with no error: a dead-looking feed that mimics segmentation falling through, so regenerate the y4m before any deeper diagnosis.

  3. Empty-scene mode (same flags, office-empty.png as the file): the guaranteed-clean no-person feed. The fixture is chosen because it does NOT false-positive (validated clean at a 0.75 segmentation-confidence cutoff; that is the fixture's property, not an app's tuned threshold), so whole-frame assertions stay unconfounded: background replacement should cover the WHOLE frame, blur should blur the whole frame uniformly, and any sharp person-shaped region that appears is the app's bug (an empty mask not honored), not the feed's. Use it for effects like blur where a person, real or spurious, would muddy the read.

Asserting the mask

Gate on page errors (collect pageerror + console error, filter favicon/DevTools noise) the same as the base, then drive the effect and Read the screenshot (vision):

  • Working mask: the person is kept (sharp, full silhouette) and the background is replaced (image effect) or blurred (blur effect).
  • Segmentation fell through: the frame is unchanged, or the WHOLE frame got the effect (no stencil). Check the network panel/console: web MediaPipe typically loads from a CDN, so an offline run or a strict CSP kills it.
  • Empty mask: the person vanishes into the background layer.
  • Halo / ragged edges: real signal, not a harness bug; this rig is how you tune the app's threshold/hardness controls. Iterate the controls and re-Read.

A static fixture cannot test temporal stability (flicker, edge crawl); for that, encode a moving subject: ffmpeg -i subject.mp4 -t 5 -r 15 -pix_fmt yuv420p -vf scale=640:480 /tmp/subject.y4m (same loop behavior).

Serving an exported static site

Use the base's serving recipe (playwright-harness, "Serving an exported static site"). The camera-specific trap: a pre-rendered page can look alive while its entry JS 404s and the camera is never requested, so a camera that never starts on the deployed site IS that bug until proven otherwise.

GPU note (delta)

WASM segmentation survives SwiftShader, so green-ball and person modes work headless with no GPU flags. Only HEAVY generative-shader layers need the base's ANGLE relaunch (--use-gl=angle --use-angle=gl); see playwright-harness, "WebGL / GPU caveat".

Scope

  • CAN validate: getUserMedia wiring, effect pipeline, mask shape, compositing, per-effect uniforms, threshold/hardness tuning by vision, the deployed site itself (point TARGET_URL at production; no install needed).
  • CANNOT validate: real FPS/perf, real-camera noise/exposure behavior, Safari/Firefox capability fallbacks (the fake-capture flags are Chromium-only), temporal mask stability from a still fixture.

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.