Vanszs/AntiSlopUI
Vanszs/AntiSlopUI has 33 GitHub stars and 7 forks. GitHot measured +2 stars in 24 hours. Last checked 2026-11-17 04:39 UTC. Last push 2026-11-14.
Measured star growth
- in 24h
- +2
- in 7 days
- Update delayed
- in 15 days
- Update delayed
- in 1 month
- Update delayed
Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.
About
Anti-AI Slop UI/UX steering skills — filter, palette, and research tool for Kiro, Claude Code, Cursor, Windsurf, Copilot, and more. Build intentional, branded, accessible interfaces.
Web & AppsREADME.md
OriginalAntiSlopUI
One skill that makes AI-generated UI stop looking machine-generated: a deterministic
ban filter, a positive pattern library, a CSV design-system generator, and an opt-in
award-grade craft tier. Everything ships in this repository — SKILL.md is the skill,
references/ is the library, ui-ux-pro-max/ is the research tool.
If you could swap the logo with a competitor and nobody would notice — it's slop.
Install
npx skills add Vanszs/AntiSlopUI
That installs one skill (antislop-design) with all references and the research tool.
Manual installs copy the whole repository, because the references and the tool live
beside SKILL.md:
git clone https://github.com/Vanszs/AntiSlopUI.git
mkdir -p ~/.claude/skills
cp -r AntiSlopUI ~/.claude/skills/antislop-design
Same shape for Codex (~/.agents/skills), opencode (~/.config/opencode/skills), and
Kiro (your-project/.kiro/skills). Do not put this under .kiro/steering/ —
steering files default to always-included and inject the entire catalog into every
request.
For agents without a skills mechanism, point them at the file:
## UI work
Read `<skill-dir>/SKILL.md` before writing or reviewing any UI code. It is a hard filter
that owns the ban list, not advice. Open the tell catalogue plus, at most, two sections of one library file per task — never a whole reference file. Run the Pre-Delivery Audit and report PASS/FAIL/N-A/UNVERIFIED per item with
evidence. Never invent social proof; omit it and record `NO-PROOF: user supplied none`.
Requirement: Python 3.8+ for the research tool and for the repository checks in scripts/.
The skill content itself — SKILL.md and references/ — is pure Markdown.
What problem this solves
LLMs generate UI by averaging training data dominated by Tailwind docs, shadcn/ui examples, abandoned repos with unchanged defaults, and 2020–2024 Dribbble SaaS shots. Output collapses toward the visual mean: purple gradients, Inter everywhere, centered dual-CTA heroes, 3-card grids, glass navbars, "Supercharge your workflow" copy.
That is structural. Every decision left unspecified gets the most probable token. The cure is deliberate, content-specific variation — enforced here, not suggested.
One skill, one contract
Version 3 merges what used to be three skills. There is no routing between packages and no precedence ladder, because everything is in one file:
| Layer | Lives in | What it does |
|---|---|---|
| Ban filter | SKILL.md |
12 hard gates, closed severity vocabulary, motion table, text measure, palette and type tables, the PASS/FAIL audit |
| Pattern library | references/ |
25 layouts, typography systems, 17 card variants, forms, nav, tables, charts, dashboards |
| Craft tier (opt-in) | references/awards-craft.md |
Awwwards-weighted rubric (Design 40 · Usability 30 · Creativity 20 · Content 10) plus hard performance and accessibility gates that deny the tier |
| Research tool | ui-ux-pro-max/ |
CSV-driven design-system generator, post-filtered before use |
Rules that used to be defensive now hold by construction: a single ban list, a single motion table, a single budget, and no dangling cross-skill steps.
The decision rule
Any P0 tell blocks delivery, as does 3+ P0/HIGH tells. MEDIUM and LOW are recorded. A tell is one catalogue row, or one of the code-level findings in the audit (five a11y/form HIGH items (focus, labels, alt, colour-only, validation) and seven code P0 items), counted per element — overlapping rows on one element collapse to the highest severity, and page-scope rows count once for the page.
WARN-BRAND-SWAP is a warning, never a block; an UNVERIFIED P0 blocks like a FAIL. N-A means the component or capability is
absent; UNVERIFIED means you could not measure it — never mark an unmeasured item PASS.
Context budget
- Always:
SKILL.md. - On demand: this file, the tell catalogue (needed for counting) and sections of one library file — never whole reference files. Ceiling: 70 KB of skill text (≈20k real tokens; 3.5 bytes ≈ 1 token).
- Never: read a whole reference file. Grep the headings — the
## Contentsblock is only a summary, grep also gives line numbers — then read that range:cd <skill-dir> && grep -nE '^#{1,2} ' references/layouts.md. - Never: read the CSVs by hand. Query them through the research tool (2–3 targeted queries).
Usage
- Filter — read the gates, the deterministic terms and the audit in
SKILL.md. - Palette — pick a pattern from
references/by its when-to-use note; vary anatomy, widths and padding. - Research — generate an industry design system, then post-filter it.
- Audit — run the Pre-Delivery Audit; then
scripts/audit.pyon the artifact in CI.
# Full design system
python3 ui-ux-pro-max/scripts/search.py "beauty spa wellness service" --design-system -p "Serenity Spa"
# Persist MASTER + page overrides into design-system/<slug>/ (gitignored)
python3 ui-ux-pro-max/scripts/search.py "fintech crypto" --design-system --persist -p "Ledger"
# Domain and stack searches
python3 ui-ux-pro-max/scripts/search.py "glassmorphism" --domain style
python3 ui-ux-pro-max/scripts/search.py "form validation" --stack react
Commands are repository-relative. When the skill is installed elsewhere, use
python3 "<skill-dir>/ui-ux-pro-max/scripts/search.py" … — the script resolves its data
directory relative to itself. Full workflow, domains, stacks and the post-filter table:
references/research.md.
What's inside
AntiSlopUI/
├── SKILL.md # the skill: gates, tables, terms, audit
├── references/
│ ├── slop-uiux-patterns.md # 126 tell rows: TELL → WHY-SLOP → FIX → SEVERITY
│ ├── slop-code-patterns.md # React/TSX/a11y code-level tells
│ ├── layouts.md # 25 numbered layout patterns
│ ├── typography.md # pairings, modular scales, clamp() recipes
│ ├── components.md # cards, buttons, forms, nav, data display
│ ├── animation.md # overlays, feedback, motion recipes
│ ├── design-systems.md # library index, tokens, decision matrix
│ ├── charts.md # chart patterns, chrome, anti-slop scan
│ ├── dashboards.md # admin shell, tables, filters, states
│ ├── awards-craft.md # opt-in craft tier, scored rubric, hard gates
│ └── research.md # CSV generator workflow + post-filter
├── ui-ux-pro-max/
│ ├── scripts/ # search.py, core.py, design_system.py
│ ├── data/ # CSV databases (query via search.py)
│ └── LICENSE # upstream MIT notice + local modifications
├── scripts/ # contributor tooling; not needed to use the skill
│ ├── validate.py # 23 repository checks
│ ├── audit.py # artifact gate (bans, glass, motion, layout)
│ ├── stats.py # recompute/splice the stats block below
│ ├── toc.py # regenerate every ## Contents index
│ ├── golden.py # semantic snapshot test over five generator fixtures
│ └── conflict-allowlist.txt # reviewed exemptions for audit/validate scans
├── tests/golden/ # committed fixtures (ASCII, markdown, MASTER, page)
└── .github/workflows/validate.yml
Verification
Claims here are checked, not asserted:
python3 scripts/validate.py # 23 structural + semantic checks, non-zero exit on failure
python3 scripts/audit.py . # artifact gate across docs and code
python3 scripts/golden.py # semantic snapshot: filtered generator output must not drift
python3 scripts/stats.py # recompute the stats block from the CSVs
python3 scripts/toc.py --write # regenerate reference indexes (non-destructive)
What is enforced: single-skill contract and frontmatter; reference/link integrity; no
stale paths; stats matching the CSVs; stack coverage; documented CLI values matching
argparse; the brand-hue table matching the generator and -p invariance; the persist
contract named in references/research.md; CSV structure (width, IDs, severity);
## Contents correctness; a single motion canon; README claim sync; and an audit of a
freshly generated MASTER.md. golden.py locks behaviour rather than structure: five
fixtures (ASCII, markdown, persisted MASTER.md, and a page override) are regenerated and
compared byte-for-byte with the timestamp normalised, so a change to the post-filter, the
ledger, the button contrast pair, or the CSVs fails CI until the fixtures are refreshed
with python3 scripts/golden.py --write.
Stats
| Metric | Value |
|---|---|
| Skills | 1 |
| Markdown files | 13 |
| Reference files | 11 |
| Reasoning rules | 100 |
| UI styles | 66 |
| Color palettes | 96 |
| Font pairings | 57 |
| UX guidelines | 99 |
| Chart types | 25 |
| Tech stacks | 13 |
Migration from v2 (three skills)
v3.0.0 merged anti-ai-slop-ui-ux, component-reference-design and ui-ux-pro-max into
this single skill. The old directories and skill ids are gone; the rule content is now
SKILL.md plus references/. If you have the old skills installed, remove them
(npx skills remove anti-ai-slop-ui-ux component-reference-design ui-ux-pro-max) so two
ban lists cannot drift apart.
Contributing
git clone https://github.com/Vanszs/AntiSlopUI.git
cd AntiSlopUI
git checkout -b feat/your-feature
# edit
python3 scripts/validate.py && python3 scripts/audit.py .
git commit -m "feat: description"
git push -u origin feat/your-feature
gh pr create
Rules of the house: no claim without a validated number, no ban without a compliant
default, no reference file without ## Contents, no motion duration outside the canonical
table, and never invent proof.
Acknowledgments
- UI/UX Pro Max by
NextLevelBuilder — the
ui-ux-pro-max/directory is their design intelligence system, vendored under MIT with local fixes (paths, CSV quoting, docstrings, a compliance post-filter in the generator). Local notice:ui-ux-pro-max/LICENSE. - Inspired by MUI, shadcn/ui, Radix, Headless UI, Material Design 3, Apple HIG, IBM Carbon, NNG, Refactoring UI, Every-Layout, Tailwind UI, Ant Design, Chakra UI, Josh Comeau, Emil Kowalski, and Adrian Krebs.
- The opt-in craft tier's weights follow the Awwwards evaluation system (Design 40 · Usability 30 · Creativity 20 · Content 10).
License
MIT — see LICENSE. UI/UX Pro Max is MIT licensed; see
ui-ux-pro-max/LICENSE.