31 October 2026 · Review-rule revision · No output test
Keep the review rules with the skill
A project documents a fixed skill commit and a separate fixed rules commit. This makes reviews easier to reproduce; no better output was tested.
30 October 2026 · Existing app UI · No output test
Review an app screen before changing it
Improve UI reports supported design findings without editing product source. Visual claims still need rendered evidence.
29 October 2026 · UI review rules · No output test
A pinned UI review skill can still use changing rules
The pinned skill fetches rules from a moving branch. Record or pin the rule revision if you need comparable findings.
25 October 2026 · Package identity · No install or output test
Same name, different frontend skill
A catalogue pointer and a full skill can use the same name. Check the installed SKILL.md and author before a result comparison.
24 October 2026 · Skill choice · No output test
Before you install a frontend skill, define the job
A one-screen brief can be the first route. Use a skill when a repeated procedure or specialized workflow justifies it. The pool has not measured which route makes a better screen.
23 October 2026 · Design-system creation · No output test
Need a design system, or just a better screen?
OpenDesigner targets reusable tokens and DESIGN.md; use existing rules for one-off screen work.
22 October 2026 · Claude Code usage check · No output test
A skill was used. Was it useful?
/skill-doctor reports usage where supported; an improved frontend still needs a no-skill comparison.
21 October 2026 · Skill-loading check · No outcome test
Installed a frontend skill? Check whether it ran.
Claude Code loads a skill body when invoked. Confirm that step before attributing an output to a skill.
20 October 2026 · Skill revision evidence · No adoption result
A 75% screenshot win rate does not settle adoption.
An open contributor PR reports better screenshot judgments for proposed text versus an older version, not a no-skill or current-source result.
19 October 2026 · Comparison method · No skill result test
Check the screen and the flow separately.
Visual fidelity and function can diverge in web-app agent tasks. A listed skill still needs a same-project, no-skill comparison.
18 October 2026 · Visual-default evidence check · Method limit
A design skill can leave a model default intact.
A contributor reported pill-shaped buttons in 10 of 21 Opus 5.5 pages using Frontend Design. No no-skill baseline or pinned skill commit was given.
17 October 2026 · Benchmark recheck · Method change
A skill result on one model may not hold on another.
Test a candidate on the model and project you use. The benchmark measured functional tasks, not visual quality or these exact listings.
16 October 2026 · Code-to-Figma source check · No change to web-skill recommendation
Need an editable Figma file, or a working web page?
Figment describes a code-to-Figma handoff route. We checked the README, not the skill body or output; it remains a watch item.
14 October 2026 · OpenDesign source check · Meaningful change
OpenDesign has two different frontend skill files.
The OpenAI-named pointer is not a full workflow; a separate Anthropic adaptation is substantive. No install or outcome test.
13 October 2026 · Game UI source check · Meaningful change
A mobile browser game is not a native mobile app
Game UI Frontend targets browser-game HUDs. Its mobile rules describe a responsive browser layout, not a native-app workflow. No skill result test.
12 October 2026 · Agent Browser source recheck · Meaningful change
Agent Browser's SKILL.md is a discovery stub
The v0.38.1 source link is a discovery stub. Load the full core workflow from the installed CLI and record its version before judging browser evidence. No runtime test.
11 October 2026 · Frontend workflow source check · Meaningful change
Need a new design, or proof the current one works?
Use concept work for an open visual target; use targeted QA for a known route and expected state. Set release checks before judging either output. No skill result test.
10 October 2026 · Anthropic skill-source check · Meaningful change
A web artifact is not your live website
Web Artifacts Builder packages a complex Claude artifact, not a live-site redesign. Start a bounded token pass in the actual site. No skill result test.
9 October 2026 · Product Design source check · Meaningful change
Know the screen before you ask for three designs.
Ideate waits for a target and user outcome before image generation. Use it for visual choice, not a small correction to an existing UI. No plugin or result test.
7 October 2026 · Trial-route source check · Meaningful change
A no-install trial still needs a complete skill prompt.
Vercel Skills documents a temporary trial; one user reported piped prompt truncation in version 1.5.23. Check what the agent received. We did not run the CLI.
6 October 2026 · Dashboard task-fit check · Meaningful change
For a dashboard, define the data task before adding a skill.
Start with an existing-system brief for the user’s decision and data. Try one conditional skill only if it fills a named gap; no result comparison was run.
5 October 2026 · Installer source check · Meaningful change
An update message does not verify your installed skill.
Vercel Skills’ inspected global-update path does not check installed bytes. Inspect local edits and copy or symlink layout after updating; we did not reproduce either user report.
Evidence log
Research reports and source checks.
28 October 2026 · Skill-source check · Meaningful change
To match a screenshot, compare the rendered screen
UXKIN ui-from-screenshot describes a reference-matching loop; no installation or output test was performed.
4 October 2026 · Candidate hold · Meaningful change
A motion default needs an off switch.
Silk Design makes smooth scrolling and reveals defaults. We held it outside the sixteen listings pending reduced-motion, mobile-readability, and rendered-result checks. One user report was not reproduced.
3 October 2026 · Skill-category check · Meaningful change
A design-system skill is a build project, not a ready-made pick.
Vercel Labs documents a source-derived skill pipeline. It belongs beside, not inside, our ready-made catalogue. Codex use and outcomes remain untested.
2 October 2026 · External benchmark check · Meaningful change
Test a web skill against no skill.
Two benchmarks found different averages under different tasks and loading setups. Neither tests our listings or visual quality.
1 October 2026 · Task-fit source check · Meaningful change
Already have a design system? Do not ask for a new palette.
The pinned Anthropic skill asks for palette and type choices. We clarified the existing-system boundary; no outcome test was performed.
30 September 2026 · Source check · Meaningful change
OpenDesign’s frontend-skill entry is not the full workflow.
OpenDesign’s entry is a discovery pointer to an OpenAI skill removed from the current upstream collection. We did not install or compare it.
29 September 2026 · Source recheck · Meaningful change
Check the React composition example before you copy it.
Verified a mismatch in the pinned lifted-state example against a related rule. The skill remains a conditional fit; no runtime result was tested.
28 September 2026 · Catalogue expansion · Meaningful change
For a form, design the failure path too.
Added one form-specific conditional skill route. The mutable source describes field necessity and error recovery; Codex loading and results were not tested.
27 September 2026 · Catalogue expansion · Meaningful change
A game HUD is not a dashboard.
Added OpenAI Game UI Frontend for browser-game HUD, menu, and overlay work that must protect the playfield and make camera or menu input boundaries explicit. Its mutable source, Codex loading, and outcome are untested by this pool.
26 September 2026 · Catalogue expansion · Meaningful change
One direction for the real screen.
Added PaulRBerg Frontend Design for an existing or substantial screen that needs one project-bound direction, local-system fidelity, and rendered inspection. Its mutable source, browser dependency, Codex loading, and outcome are untested by this pool.
25 September 2026 · Catalogue expansion · Meaningful change
Choose a visual direction before you build.
Added OpenAI Product Design Ideate for three image-based directions before prototype work. It needs plugin availability, a minimum brief, references, Image Gen, and a human choice. The mutable source and its outcome are untested by this pool.
24 September 2026 · Adoption gate · Meaningful change
A detailed frontend workflow is not ready if its package metadata fails.
Held Snowe UI Skill outside the catalogue because GitHub currently reports a YAML error in its SKILL.md frontmatter. We did not install it or test host behavior. Check package readiness before workflow fit.
23 September 2026 · Catalogue expansion · Meaningful change
Make the visual decision before you ask Codex to apply it.
Added source-pinned Wenkang Frontend Design for a real screen or defined concept that needs human visual choices, element annotations, and an exported decision packet before implementation. The local preview is not proof that the live route matches. No pool run was performed.
22 September 2026 · Catalogue expansion · Meaningful change
A design skill needs a complete state model.
Added source-pinned PracticalSwan Frontend Design 2.0 as a conditional path for substantial product work that needs explicit system context, complete states, responsive checks, and rendered verification. Its protocol is author guidance, not a pool result test.
21 September 2026 · Catalogue expansion · Meaningful change
QA needs the right browser surface.
Added OpenAI Frontend Testing & Debugging for a named rendered change or bug flow. Its source is mutable, and QA evidence now records backend, viewport, session context, target URL, and visible state. No pool run was performed.
20 September 2026 · Catalogue expansion · Meaningful change
Dogfood the flow. Do not ask it to invent the design system.
Added Agent Browser Dogfood 0.38.1 for exploratory QA of a runnable frontend when a browser finding needs replayable evidence. It does not read source, enforce a design system, or certify that an interface is good. No pool run was performed.
19 September 2026 · Catalogue expansion · Meaningful change
A browser skill can show the state. It cannot judge the design.
Added Vercel Agent Browser 0.38.1 for a runnable UI state that needs navigation, snapshots, and retained capture. It can create an evidence route for later review, not a design or accessibility verdict. No pool browser run was performed.
18 September 2026 · Catalogue expansion · Meaningful change
A screenshot audit needs a real flow.
Added OpenAI Product Design Audit for a product journey you can capture and inspect. It is not a substitute for source review, and screenshots cannot establish full accessibility compliance. No pool audit was performed.
17 September 2026 · Catalogue expansion · Meaningful change
A visual skill needs a visual proof path.
Added Frontend App Builder for a new high-visibility visual surface when you can approve a concept and compare a browser render. No pool run was performed.
16 September 2026 · Adoption guidance
A portable skill can still be the wrong project configuration.
Added an adoption gate: target agent, scope, source pin, local edits, and update review.
What comes next
Run a disclosed existing-system screen against a careful no-skill brief. Compare token and component reuse, required states, and the rendered result. The local comparison capability is still pending.