- Home
- Skills
- Testing & Debugging
- Design QA: Fidelity Check
Design QA: Fidelity Check
Measure how close your HTML/React prototype is to the reference, per viewport, and get a pass / iterate verdict that says what to fix.
$12
Works with the AI Agents you already use
Design QA: Fidelity Check
Example session with this skill installed
Check my pricing page prototype (http://localhost:5173/pricing) against the reference https://example.com/pricing on iOS and desktop. Is it close enough, and what should I fix first?
- Read your context and instructions
- Compiled the design qa
- Generated the UI component
Captured the reference and the candidate with the same frozen settings (DPR 1, animations off, fonts ready), measured both viewports and ran the verdict in faithful mode.
Overall: ITERATE (web passes, ios needs work)
| viewport | edge_f1 | phash | palette | fidelity_index | contrast | judge | verdict |
|---|---|---|---|---|---|---|---|
| ios 390x844 | 0.71 | 0.82 | 0.95 | 0.88 | 2 nodes | n/a | iterate |
| web 1440x900 | 0.93 | 0.91 | 0.98 | 0.96 | 0 | n/a | pass |
Failed gates on ios
- G4 edge_f1 0.71 < 0.75 (threshold band 0.75-0.90) -> class_hint: typography
- G2 axe contrast: 2 nodes fail WCAG AA (
.plan-card .price-note,.faq summary)
Top 3 fixes
- Pricing cards on mobile: the plan title wraps to 2 lines. The reference uses 20px/28px, the candidate 24px/32px. Match the type scale in
PlanCard.tsx. - Raise contrast of
.price-note(#9CA3AF on white, 2.5:1) to at least 4.5:1, e.g. #6B7280. - Card gap is 24px in the candidate vs 16px in the reference (see the red band between cards in the diff).
Files
.design-qa/pricing/runs/20261114-1520-1/metrics.json,verdict.json.design-qa/pricing/runs/20261114-1520-1/diff@390x844.png(red = changed pixels, blue = masked status bar)- History appended:
.design-qa/history.md(pricing: run 1, best = web pass)
Next: apply the 3 fixes, capture into a new run folder and re-check ios.
PlanCard.tsx
TSX · React component
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
"Looks close enough" is not a review. This skill has your agent capture the reference (live URL, local HTML, phone screenshot or Figma/PDF export) and your prototype under identical frozen browser settings, for iOS 390x844, Android 412x915 and desktop 1440x900.
What it measures
Each viewport gets SSIM, pixel diff, perceptual hash, edge F1, a dominant-palette distance, a DOM token and layout diff, and axe-core contrast.
Explained verdicts
Gates (G1 invariants, G2 contrast, G3 layout, G4 edges, G5 palette, G6 fidelity index, G7 optional judge) turn the numbers into a verdict: pass, iterate, grey, fail, incomplete or uncalibrated. Each verdict comes with a class hint (colour, typography, layout, copy, shift) and diff images, so the agent knows what to change next. Viewports are never averaged.
Modes
- Faithful (default): match the reference
- Brand: composition from the reference, colours checked against your own design tokens
- Pairwise: A/B, which variant is closer or better
- Regression: same design before/after a refactor
A local history (.design-qa/history.md) tracks progress across iterations. All thresholds and metrics are documented, and calibration data ships with the skill.
What you get
SKILL.md: a 10-step workflow from capture to verdict and reportscripts/: snapshot, fidelity, verdict, palette check, phone-screenshot prep, token diff, judge merge, run history and optional local re-calibration (Python)scripts/params.default.json(starter thresholds) andscripts/calibration.json(synthetic calibration, no external references)references/: what each metric can and cannot see, the gate and verdict logic with a fix brief per change class, and workflow recipes (dev server vs Figma export, phone screenshot, dark mode, brand mode, A/B, regression)
Permissions / network
- Opens only the URLs and files you give it, plus the sub-resources those pages load.
--block-externalrestricts this to local and same-origin requests - Install step contacts PyPI, the Playwright browser CDN and optionally npm. No telemetry, no uploads
- Writes only to
.design-qa/in your project (configurable withDESIGN_QA_DIR)
Part of the Design QA set with Reference Capture, Designer Judge and Device Frame. Each skill works on its own.
Reference and prototype must be rendered on the same machine (cross-OS font rendering never matches). Starter thresholds come from synthetic pages, so tune them for your product. A PNG reference has no DOM, so token diff and reference contrast are unavailable. Not for native mobile code.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 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 day ago
Requires
Permissions
File Scopes
Terminal runs the bundled Python scripts. Browser/Network: headless Playwright Chromium opens only the URLs and files you provide (and their sub-resources); --block-external keeps it local. Install contacts PyPI, the Playwright CDN and optionally npm. Writes only to .design-qa/. Optional env vars DESIGN_QA_DIR and AXE_JS change paths only; no secrets. A local git rev-parse tags runs.