design-loop
Skill1 starsChanged 53 days agoMITView on GitHub
Visually judge a page or section of a web-kit site against its DESIGN.md and gold-standard reference, then fix it. Captures screenshots at desktop and mobile widths in every language, scores them with a strict rubric, and fixes the biggest gaps. Repeats until the page passes. Use it after building or changing any visible section, when the user says something "looks generic", "looks AI-made", "no me convence", "pulir el diseño", or "compare with the reference", and in the pilot and launch phases of web-kit:new-site.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- seankoji-com/.agentstheir own version, first seen Nov 8, 2026
- tstapler/dotfilestheir own version, first seen Oct 10, 2026
- 17-sss/agent-skillstheir own version, first seen Sep 22, 2026
- SkotAlexsander/skillstheir own version, first seen Sep 16, 2026
- saint0x/my-skillstheir own version, first seen Aug 29, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- tstapler/dotfiles★ 8same name
- 17-sss/agent-skills★ 2same name
- seankoji-com/.agents★ 2same name
- saint0x/my-skills★ 1same name
- SkotAlexsander/skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository7 other files it gives its agents.
- template/CLAUDE.mdCLAUDE.md
- assetsSkill
- design-systemSkill
- new-siteSkill
- publishSkill
- retroSkill
- site-copySkill
The file
We split this skill into the rules it gives an agent
4 rules. 0 are common, 1 is distinctive, and 3 are very specific to this project.
0
Common
Most of the 5 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 12
Start the dev server in the background (npm run dev) and note the URL.
- Very specific: line 19
Deterministic checks: Run node <kit>/scripts/check.mjs design <files you touched>, where <kit> is <base>/../...
- Distinctive: line 21
Pass: when every item scores ≥ 4, the deterministic checks are clean, and there are no console errors.
- Very specific: line 24
Stop after 5 iterations even if the page hasn't passed.
“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Design loop
The first version is never the final one. This loop replaces "looks fine to me" with a comparison against a real reference. The reference is the site named in DESIGN.md → Sources, or a screenshot in docs/references/.
Setup
- Start the dev server in the background (
npm run dev) and note the URL. - Load the browser tools in one call:
ToolSearchwithselect:mcp__claude-in-chrome__tabs_context_mcp,mcp__claude-in-chrome__tabs_create_mcp,mcp__claude-in-chrome__navigate,mcp__claude-in-chrome__computer,mcp__claude-in-chrome__resize_window,mcp__claude-in-chrome__javascript_tool,mcp__claude-in-chrome__read_console_messages. If the extension doesn't respond, ask the user to connect it. Don't judge design from code alone. - Open a new tab with the page, and another with the reference.
Each iteration
- Capture. Screenshot the section at 1440×900 and at 390×844, in every language (click the language toggle). Also screenshot the reference at the same widths, once per loop. Check the console for errors.
- Deterministic checks. Run
node <kit>/scripts/check.mjs design <files you touched>, where<kit>is<base>/../... Any finding counts as a failure. - Score every rubric item in rubric.md from 1 to 5. Write one line of evidence for each score, based on what you see in the screenshots. Score the way a studio art director would when your work is next to the reference: a 5 means indistinguishable in craft from the reference, not "acceptable".
- Pass when every item scores ≥ 4, the deterministic checks are clean, and there are no console errors.
- Otherwise pick the 3 biggest gaps (lowest score, then most visible) and fix them in code. Fix root causes: a token, a component recipe, spacing on a shared utility. Don't add per-element patches. If the fix is a design decision that
DESIGN.mddoesn't cover, record the decision there first.
Stop after 5 iterations even if the page hasn't passed. Report the remaining gaps and what each one needs. Often the gap is an asset (web-kit:assets) or a decision the user must make.
Report
Output a short table (rubric item, score, evidence) plus the fixes you made. In the pilot, also show the final screenshots to the user.
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.