liquid-glass-debug
Skill0 starsChanged 7 days agoMITView on GitHub
Use when glass works in Chrome but is broken on iPhone, bend/refraction is missing in Safari, rim looks painted or like a border, glass looks like plastic, backdrop-filter url() paints nothing, edges are square, nested glass is milky, or optical measurements must be verified.
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- docs/llms.txtllms.txt
- liquid-glass-suiteSkill
- liquid-glass-nativeSkill
- liquid-glass-opticsSkill
- liquid-glassSkill
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
- Very specific: line 8
Start with the symptom, then use the full first checks and diagnostic order in ../../shared/references/11-failure-catalogue.md.
- Very specific: line 50
Strategy/parity: use liquid-glass-optics for WebGL2, per-geometry maps or reference matching.
- Distinctive: line 54
Run from the suite root. Map/chroma scripts use the Python standard library.
- Distinctive: line 72
Test the actual target engine and the explicit fallback.
- Distinctive: line 73
Verify reduced transparency, reduced motion, forced colors, corner geometry and the built bundle after the last edit.
≈ “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.
Files
Liquid Glass debug
Start with the symptom, then use the full first checks and diagnostic order in ../../shared/references/11-failure-catalogue.md. Do not tune tokens before checking chain delivery, wiring, filter region and map alpha.
| # | Symptom / entry | First move |
| S01 | Just blur, not glass | Check live url() and edge/interior structure |
| S02 | Frosted plastic | Put url() last; remove blur after it |
| S03 | Rim looks like a painted/dirty border | Remove rim blur; use mask + inset treatment |
| S04 | White rim on white panel | Measure contrast ceiling; change structure |
| S05 | Shine is a static stripe | Use size-relative rim thickness |
| S06 | Same shine on every panel | Check where clamp() stops binding |
| S07 | Shine does not track pointer | Verify delegated flare listener |
| S08 | Nearest edge does not brighten | Check distance ramp and eased boost |
| S09 | Transparent even at edges | Find a zero-floored feather mask |
| S10 | Bottom/some corners square | Clip rectangular children |
| S11 | Curved panel, square blur | Restore exact filter region; widen map ramp |
| S12 | Dark glass looks flat/plastic | Check opacity, shadow, then dim law |
| S13 | Practically black | Check for affine/flat-multiply dim |
| S14 | Flat on the ground | Measure backdrop luminance range |
| S15 | Rim present, bend lost | Check computed url() and live filter IDs |
| S16 | Thicker rim made it worse | Inspect hard-cutout mask |
| S17 | Still lines, now thicker | A ring cannot fade inward |
| S18 | Change made it flatter | Diff mask, shadow and bend layers separately |
| S19 | Something indescribably off | Fix a starved backdrop field first |
| S20 | Line behind panels | Prove transmitted content vs painted line |
| S21 | Logo disappeared | Check per-theme color-math filters |
| S22 | Logo is just a circle | Preserve raw computed token before transforms |
| S23 | Menu closes before shape returns | Fit child exit inside container budget |
| S24 | Ghost blur after close | Never fade opacity on filter carrier; split veil |
| S25 | Cards resize/jelly on scroll | Remove content-visibility:auto |
| S26 | Panels lag scroll | Check compositor vs rAF clocks |
| S27 | Pointer is a flashlight | Size falloff in px against rim thickness |
| S28 | White stripe through body | Delete the diagonal glint layer |
| S29 | Washed-out/milky panel | De-nest glass; stop stacking fills |
| S30 | Chrome works, iPhone broken | SVG backdrop is WebKit-blocked; keep fallback or use WebGL2 |
| S31 | Bend/refraction not showing | url() last, map alpha 255, output not raw map |
| S32 | Bend is not curved | Check square map and oversized filter region |
| S33 | Rim and panel corner disagree | Rebuild map from painted corner geometry |
Engine quirks
- Chromium: alpha-0 maps die; filters cache by ID; backdrop ends at the element edge; transforms can skew the lens. See
../../shared/references/05-chromium-quirks.md. - Safari/WebKit:
backdrop-filter:url()is blocked in documented WebKit behaviour (245510); contentfilter:url()is different. Filter IDs cache, promoted children can leave an ancestor filter, and Playwright WebKit is not a real iPhone — the owner's physical iPhone Safari renders the standalone showcase's bend (2026-11-08; verify on target versions). See../../shared/references/06-safari-webkit.md. - Strategy/parity: use
liquid-glass-opticsfor WebGL2, per-geometry maps or reference matching. A syntax@supportsresult is not paint evidence.
Measure
Run from the suite root. Map/chroma scripts use the Python standard library; edge profiles require Pillow.
python shared/scripts/measure-displacement-map.py tests/fixtures/lens-map.png --scale 40
python shared/scripts/edge-profile.py tests/fixtures/rim-swatch.png --at 0.45 --span 40
python shared/scripts/chroma-retention.py --in 240,232,210 --sat 1.4 --wash 0.30 --tint 31,29,41
python -m unittest discover -s tests -v
node scripts/build-showcase.mjs --checkFor app maps, replace the fixture path and declared scale; for rim/glow/cutout comparisons use the corresponding fixture. Commands, live-bundle audits and thresholds are in ../../shared/references/10-measurement.md. Run npm test only when Playwright browsers are installed; otherwise report browser behavior untested.
Verification gate
From ../../shared/references/15-honesty-and-checklists.md:
- Capture a large panel over a fine grid; small screenshots hide edge/region defects.
- Over ruled stripes, prove compression/flipping at the edge; include a zero-effect control.
- Test the actual target engine and the explicit fallback; content-filter success does not prove SVG-backdrop support.
- Verify reduced transparency, reduced motion, forced colors, corner geometry and the built bundle after the last edit.
- Report renderer, browser, geometry, settings and actual output. Never claim an unrun check, real-device pass or visual parity.
Part of liquid-glass-suite — install the suite, not this folder alone.
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
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.

