Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    UI Screenshot Drift Gate

    1

    Stop silent UI drift. Compare two local PNG screenshots, highlight changed regions, and get a reproducible PASS/FAIL gate with evidence.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    UI Screenshot Drift Gate

    UI Screenshot Drift Gate

    Example session with this skill installed

    Run UI Screenshot Drift Gate on the bundled synthetic examples/baseline.png and examples/candidate.png. Apply examples/masks.json with tolerance 8 and a 0.5% maximum changed proportion. Return the measured verdict, report locations and diff PNG; do not capture screenshots or change source files.

    • Read your context and instructions
    • Compiled the ui screenshot drift
    • Generated the UI component

    UI Screenshot Drift Gate — Synthetic Demo

    Verdict: FAIL — the candidate exceeds the configured visual-change budget. This is a measured result from the synthetic PNG examples bundled with the skill, not a customer deployment.

    MeasurementValue
    Image size800 × 480
    Evaluated pixels380,100
    Explicitly masked pixels (fictional timestamp)3,900
    Changed pixels2,998
    Changed proportion0.7887%
    Maximum accepted proportion0.5%
    Per-channel tolerance8 / 255
    Changed bounding boxx=531, y=352, width=202, height=49

    Largest changed tiles

    Tile originChanged pixels
    (512, 320)793
    (704, 320)703
    (576, 320)424
    (512, 384)374
    (704, 384)373
    (640, 320)330
    (640, 384)1

    The synthetic fixture contains a blue action button shifted by 22 pixels; that fact comes from the fixture generator, not image classification. The gate creates report.md, report.json, and a red-highlighted diff.png under the chosen output directory.

    Review before release

    1. Inspect the diff PNG and high-change tiles to decide whether the shift was intentional.
    2. Confirm stable browser, viewport, device pixel ratio, fonts and scroll state.
    3. Check that every exclusion mask is authorized and narrowly scoped.
    4. Run separate functional, accessibility and responsive tests.

    Interpretation: FAIL means this pair exceeds the pixel drift threshold. PASS would not prove the UI works, looks good or satisfies accessibility requirements.

    diff.png

    TSX · React component

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Your coding agent says the UI is finished. Did the layout actually stay intact?

    UI Screenshot Drift Gate compares two user-supplied, identically sized 8-bit PNG screenshots locally. A dependency-free Python script measures visual changes, highlights drift in red, and writes review-ready report.md, report.json, and diff.png files.

    Included

    • Deterministic per-pixel changed-rate gate with configurable tolerance and an explicit acceptance threshold.
    • Highlighted diff image, changed bounding box, and top hotspot tiles for fast review.
    • Optional user-approved rectangle masks for truly dynamic regions; fully masked images are blocked.
    • Strict input checks for PNG CRC, size limits, matching dimensions, malformed masks, and no silent overwrites.
    • Full synthetic before/after UI example, masks, measured report, references and self-tests.
    • No API key, network, Node, Playwright, subscription, external dependency, or creator access to your files needed for the Python gate.

    Best for

    Verifying before/after snapshots of AI-edited web and app UI, triaging unintended layout shifts, and preparing evidence for release review.

    Important limitations

    The skill does not capture screenshots, browse pages, judge aesthetics or functionality, scan accessibility, or auto-fix code. Supply stable local PNG pairs captured at matching viewport, DPR, browser and fonts. PASS means only that this pair meets the chosen visual drift threshold. Other formats must be exported to supported 8-bit non-interlaced PNG.

    Requires two user-authorized, identically sized, 8-bit non-interlaced PNGs captured with the same browser, viewport, DPR, scroll state and fonts. The skill does not capture images, browse sites, judge aesthetics, test functionality or accessibility, or fix code. PASS only means this image pair stayed under the chosen pixel-change threshold.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    Requires

    Python 3.9+ (standard library only)
    Python

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    Terminal: run the offline Python comparison. Read: only the user-selected baseline/candidate PNGs and optional mask JSON. Write: only the user-selected local report output directory. No browser, network or environment variable access is required.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    Not to run the diff. You need two pre-captured local PNG files; obtaining them is a separate workflow.

    The fraction of changed unmasked pixels is within the configured threshold. It does not establish that the UI looks good or works properly.

    Yes, one matching-size pair at a time. The two inputs in each run must have exactly the same dimensions.

    Yes, with narrow, explicitly reviewed JSON rectangle masks. The output reports every excluded pixel.

    The included script makes no network calls and only writes local reports. Your AI-agent host may have its own data policies.

    Non-interlaced 8-bit grayscale, RGB and RGBA PNG variants. Re-export palette or high-bit-depth sources first.

    New in Testing & Debugging