house-style
Skill2 starsChanged 5 months agoMITView on GitHub
Extract an opinionated design house-style from a reference (URL/screenshot/Dribbble) into concrete tokens + an HTML template, then apply it, so UI stops looking generic/"vibe-coded". Use when building or fixing any website, landing page, deck, doc, or web artifact, or when a UI reads as default Material/Bootstrap/system-sans. Vendored from meefs/style-skills.
10 other projects use a version of this.
Adopted by
Most recently taken up by
- koaeraser/ARMStheir own version, first seen Nov 15, 2026
- enstw/thesisbug2their own version, first seen Nov 8, 2026
- RightBlogger/bloggingskillstheir own version, first seen Nov 1, 2026
- thescottking/b2b-pmm-skillstheir own version, first seen Oct 25, 2026
- CarriedWorldUniverse/nexustheir own version, first seen Oct 12, 2026
- samuelgursky/davinci-resolve-mcptheir own version, first seen Oct 8, 2026
Other versions10 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 10
- samuelgursky/davinci-resolve-mcp★ 4.1ksame name
- ajsahni/agents-office★ 601same name
- koaeraser/ARMS★ 40same name
- RightBlogger/bloggingskills★ 4same name
- KodingDev/skills★ 2same name
- thisisdhika/skills★ 1same name
- CarriedWorldUniverse/nexus★ 0same name
- enstw/thesisbug2★ 0same name
- thescottking/b2b-pmm-skills★ 0same name
- ztrange/claude-skills★ 0same 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.
- classifySkill
- critical-qaSkill
- literature-loopSkill
- marginal-contributionSkill
- md-to-pdfSkill
- ship-webSkill
- trp2Skill
The file
Files
House Style — kill the regression to the mean
The problem it solves (meefs/style-skills): unguided, an AI regresses to the mean of its training data — mostly component-framework UI — so "make a webpage" keeps producing the same rounded, shadowed, system-sans look. That is the "vibe-coded" tell, and people ignore it. The corrective is not more polish. It is imposing ONE opinionated canon instead of the statistical center.
Workflow (do this, in order)
- Pick a reference that embodies the target aesthetic — a live URL, a clean full-page screenshot, or a Dribbble shot. Sharper reference = sharper extraction.
- Run the decomposition prompt (below) against it → get a
SKILL.mdwith concrete tokens + a self-contained HTML template. Gets you 80-100% of the way. - Generate, compare, refine. Produce a real page in the new style, hold it against the original, feed drift back in plain language ("accent #15706a, cap body at 66ch, headings in Fraunces 900"), regenerate. One or two passes closes the last 20%.
- Save the extracted style as its own
SKILL.mdunder~/.claude/skills/<style-name>/so it is reusable and every page stays consistent.
THE DECOMPOSITION PROMPT (paste verbatim, attach/link the reference)
You are a design-systems analyst. I'll give you a design reference —
a screenshot, an image, or a URL (for example a Dribbble shot).
Decompose it into a reusable "house-style" skill.
STEP 1 — Extract the canon (the recurring, defining traits):
• Palette — every distinct colour as a hex value, named by role
(bg, ink, accent, muted, line…); include any variants.
• Type — font families by role (display / body / mono), their
weights, the size scale, line-height and letter-spacing.
• Layout — content width and line length (the measure), spacing
scale, grid, border weight, corner radius, elevation.
• Signature — the 3–5 traits that make the look recognisable, plus
the anti-patterns that would break it.
STEP 2 — Emit the skill:
1. name — a short kebab-case identifier.
2. description — one or two sentences: what it makes, when to use it.
3. Design system — the tokens and rules above as concrete, non-
negotiable values (CSS custom properties for colour;
exact px / ch / rem for everything else).
4. Template — one complete, self-contained HTML document that
already renders the look, ready to clone and refill.
Output as a SKILL.md file with YAML frontmatter. Keep every value
specific; where the reference is ambiguous, choose the most
defensible option and note the assumption.The canon (pick by communication goal, not by taste)
Deep/enduring (★ favor these): Swiss/International (grid, sans, objective — reports, wayfinding) · Bauhaus · Constructivism · De Stijl/Mondrian · Art Deco (geometric luxury) · Editorial/Magazine (humanist serif, warm paper, the measure — guides/playbooks). Digital-era (trend-bound, deploy with awareness): Neo-brutalism (monospace, hard borders — dashboards) · Terminal/TUI (mono on black — dev tools/status) · Material/Flat/Glassmorphism/Neumorphism/Skeuomorphism. Mood: Minimalism (Rams, "less but better" — product pages) · Synthwave/Vaporwave · Maximalism · Tufte/Analytical (data-ink max — dashboards/metrics).
Principles: regression-to-the-mean is the default enemy · canon forces opinionated results · the measure (line length) is a foundational token, not an afterthought · every style maps to a communication goal.
Recommended house style for NOESIS (a credible, anti-casino consensus protocol)
Positioning = rigorous, technical, contribution-not-wealth, NOT hype. Recommended spine: Swiss/International grid + Minimalism (Rams) for the landing/app (signals rigor + restraint = credibility), Editorial for longform (Manifesto, whitepaper), Tufte/Terminal for the reference-node metrics/status pages. This is coherent, opinionated, and the structural opposite of the generic-Material "vibe-coded" look. Aesthetic is Will's call; decompose a chosen reference before building.
Credit: meefs/style-skills (https://github.com/meefs/style-skills). Shared by Meefs 2026-07-02.
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.

