web-taste
Skill0 starsChanged 6 months agoMITView on GitHub
Anti-slop web frontend skill — AUTO-TRIGGER mọi web UI task. Force ship Kimi/Lovable-tier quality bằng 3 luật cứng: (1) Turn 1 emit Discovery Form lock brief TRƯỚC khi gen pixel (cho task lớn) hoặc apply quality bar (cho tweak nhỏ), (2) Turn 2 pick 1 trong 5 visual directions deterministic (OKLch palette + font + posture) hoặc extract brand-spec — KHÔNG freestyle, (3) Turn 3+ TodoWrite plan → live updates → 27 anti-pattern checklist + 5-dim critique + quality tier scoring → ship single artifact. AUTO-TRIGGER khi: chạm file .tsx/.jsx/.ts/.vue/.svelte/.astro/.html/.css/.scss/.tailwind, hoặc user nói "build/sửa/render/design/tạo/refactor" + "page/landing/UI/component/storefront/admin/site/web/trang/giao diện/style/theme", hoặc "design như Kimi/Lovable/Linear/Stripe/Apple/Monocle/Vercel/Notion". Bao gồm: 9 references (directions/discovery-form/anti-slop/critique/design-md-format/quality-tiers/hero-patterns/section-blueprints/motion-recipes), 8 starter design systems (Kimi, Lovable, Linear, Stripe, Apple, Monocle, Brutalist, Datadog). Distilled từ open-design (24K stars), taste-skill, impeccable.style, getdesign.md, kimi.com, lovable.dev, huashu-design, guizang-ppt. KHÔNG dùng cho: mobile (Flutter/iOS/Android/RN — dùng mobile-design), backend (API/DB), config/infra (CI/env/docker).
3 other projects use a version of this.
Adopted by
Most recently taken up by
- ssaarthakk/dev-setuptheir own version, first seen Nov 8, 2026
- sonhaicoder/web-taste-skillword for word, first seen Nov 8, 2026
- lllkn/web-tastetheir own version, first seen Oct 5, 2026
Other versions3 projects use a version of this.
Word for word · 1
- sonhaicoder/web-taste-skill★ 2100% same
Edited · 0
None found.
Wrote their own version · 2
- lllkn/web-taste★ 4same name
- ssaarthakk/dev-setup★ 2same 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 repository12 other files it gives its agents.
- backend-fastapiSkill
- brief-recapSkill
- code-reviewSkill
- debug-firstSkill
- deploySkill
- flutter-architectureSkill
- frontend-designSkill
- git-prSkill
- harnessSkill
- kimi-renderSkill
- lme-flutterSkill
- mobile-designSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 200
THE LILA BAN: No purple/violet glow, no neon gradient, no "AI Purple" aesthetic
- Very specific: line 304
1-3: No automatic animation. CSS :hover/:active only
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Web-Taste — Anti-Slop Frontend Skill (Kimi/Lovable-tier)
Mục tiêu: ship UI đạt Kimi-tier hoặc Lovable-tier — senior designer level, không phải "đẹp tạm được". Triết lý: AI không freestyle. Lock brief → pick deterministic direction → execute với guardrails + ready-to-paste blueprints → ship 1 artifact. Source: distilled từ open-design (24K stars), taste-skill, impeccable.style, getdesign.md, kimi.com, lovable.dev.
1. KHI NÀO ACTIVATE — AUTO-TRIGGER MỌI WEB TASK
AUTO-TRIGGER khi (BẤT KỲ điều kiện nào):
✓ Chạm file: .tsx, .jsx, .ts, .vue, .svelte, .astro, .html, .css, .scss
✓ Chạm thư mục: components/, pages/, app/, src/, public/
✓ User nói "build / sửa / render / design / tạo / refactor / fix / thêm" +
"page / landing / UI / component / storefront / admin / site /
web / trang / giao diện / style / theme / form / modal / card"
✓ User mention brand: "design như Kimi/Lovable/Linear/Stripe/Apple/
Monocle/Vercel/Notion/Cursor/v0/Anthropic"
✓ User attach Figma URL / screenshot / mockup
✓ Project có package.json với react/vue/svelte/next/astro
DEPTH OF APPLICATION (skill scale theo task size):
┌─ Big task (build từ scratch / new page / major refactor)
│ → FULL workflow: Discovery Form → Direction → TodoWrite → 9 steps → tier scoring
│
├─ Medium task (refactor section / theme rebuild / new feature UI)
│ → SKIP Discovery Form, apply: directions + anti-slop + 5-dim critique
│
└─ Small task (tweak / bug fix UI / 1 component edit)
→ APPLY: anti-slop checklist + tactile press + responsive check
→ KHÔNG mở Discovery Form (waste user time)
KHÔNG dùng khi:
✗ Mobile app (Flutter/RN/iOS/Android/SwiftUI/Compose) → mobile-design
✗ Backend (API/DB/service/migration) → backend skills
✗ Config/infra (CI/CD/env/docker/deploy) → no UI involved
✗ Pure logic/algorithm (no UI surface) → general coding skill2. BA LUẬT CỨNG (KHÔNG ĐƯỢC PHÁ)
LUẬT 1 — Turn 1 PHẢI emit Discovery Form, KHÔNG code ngay
Khi user mở task mới hoặc brief mới, output đầu tiên là 1 dòng prose ngắn + Discovery Form. KHÔNG file read, KHÔNG Bash, KHÔNG TodoWrite, KHÔNG extended thinking.
Tại sao: form là time-to-first-byte. User nhanh ở radio, chậm ở redirect sai. 30 giây radio beats 30 phút redirect.
Form template:
{
"title": "Quick brief — 30 giây",
"description": "Lock những cái này trước khi build. Skip cái không cần — fill default.",
"questions": [
{ "id": "output", "label": "Build cái gì?", "type": "radio", "required": true,
"options": ["Landing/marketing page", "App prototype", "Dashboard / tool UI", "Editorial page", "Slide deck", "Khác"] },
{ "id": "platform", "label": "Surface chính", "type": "radio",
"options": ["Mobile (375px first)", "Desktop web", "Responsive all sizes", "Fixed canvas 1920x1080"] },
{ "id": "audience", "label": "Cho ai?", "type": "text",
"placeholder": "VD: nhà đầu tư seed, dev-tools buyer, exec review nội bộ" },
{ "id": "tone", "label": "Visual tone (≤2)", "type": "checkbox", "maxSelections": 2,
"options": ["Editorial/magazine", "Modern minimal", "Tech/utility", "Luxury/refined", "Brutalist/experimental", "Soft/warm", "Playful/illustrative"] },
{ "id": "brand", "label": "Brand context", "type": "radio", "required": true,
"options": ["Pick a direction for me", "I have brand spec — share file", "Match reference site/screenshot"] },
{ "id": "scale", "label": "Quy mô", "type": "text",
"placeholder": "VD: 1 hero + 4 sections, 8 slides, 4 mobile screens" },
{ "id": "constraints", "label": "Khác?", "type": "textarea",
"placeholder": "Real copy, font bắt buộc, thứ phải tránh, deadline…" }
]
}Quy tắc form:
- JSON valid (no comments, no trailing commas)
- ≤7 questions, batch 2 nếu cần
- Adapt theo brief — drop default user đã trả lời
- Lead 1 dòng prose ngắn ("Got it — landing cho SaaS B2B. Tell me the rest:")
- Sau form → STOP. KHÔNG narrate.
CHỈ skip form khi:
- User reply bên trong active design với tweak ("headline to hơn", "đổi ảnh slide 3")
- User explicit nói "skip form", "just build", "no questions"
- Message bắt đầu bằng
[form answers — …]
LUẬT 2 — Turn 2 branch theo brand
Branch A: brand: "Pick a direction for me"
KHÔNG TodoWrite. Emit form thứ 2 với 5 visual directions (xem references/directions.md):
[direction-cards form]
├── editorial-monocle — print magazine, serif, off-white + warm rust
├── modern-minimal — Linear/Vercel, system fonts, near-greyscale + cobalt
├── warm-soft — Stripe pre-2020, cream bg + terracotta + serif display
├── tech-utility — Datadog/GitHub, dense data, mono numerics + signal green
└── brutalist-experimental — Are.na/Yale, loud serif + monospace body + hot redUser pick → bind palette + font VERBATIM vào :root. KHÔNG improvise OKLch values.
Nếu user fill accent_override → dùng accent đó, giữ rest direction defaults.
Branch B: brand: "I have brand spec" hoặc "Match reference"
Brand-spec extraction TRƯỚC TodoWrite (5 steps):
- Locate — list attached files / WebFetch
<brand>.com/brand|press|about - Download — CSS, brand-guide PDF, screenshots
- Extract —
grep -E '#[0-9a-fA-F]{3,8}'lấy hex; eyeball screenshots typography. KHÔNG đoán từ memory. - Codify — Write
DESIGN.mdở project root (xemreferences/design-md-format.mdcho format) - Vocalise — state system trong 1 câu để user redirect sớm
Sau đó chuyển LUẬT 3.
Branch C: anything else / no brand
Skip thẳng LUẬT 3 với tone-based default direction.
LUẬT 3 — TodoWrite plan, live updates, ship 1 artifact
Sau khi direction/brand-spec lock → first tool call là TodoWrite với 5-10 imperative items:
1. Read DESIGN.md (brand-spec hoặc chosen direction)
2. Bind palette/font vào :root
3. Plan section/screen/slide list với rhythm (state TRƯỚC khi viết)
4. Wireframe pass — grey blocks + labelled placeholders, ship visible early
5. Fill real copy từ brief (replace [REPLACE] placeholders)
6. Fill imagery / decorative (real screenshots, not stock)
7. Self-check: chạy 27-anti-pattern audit (references/anti-slop.md, P0 PHẢI all pass)
8. 5-dim critique: philosophy/hierarchy/execution/specificity/restraint, fix < 3/5
9. Emit single <artifact>Live update rules:
- Mark step
in_progressTRƯỚC khi start - Mark
completedngay khi xong (KHÔNG batch ở cuối turn) - Plan đổi → edit list, đừng silent abandon
Step 7-8 KHÔNG NEGOTIABLE. Failed P0 → fix trước khi emit.
3. MODE — BRAND vs PRODUCT (impeccable.style)
Trước khi viết CSS, chọn mode dựa vào output type:
BRAND MODE Marketing/editorial/landing/pitch deck
→ Emotion-driven, hero-led, 1 decisive flourish, real photography
→ Display font có character (serif hoặc display-sans), tracking-tight
→ Whitespace generous, long-form sections OK
→ Animation: orchestrated reveals, parallax, cinematic
→ AI Tells to avoid: filler copy, fake metrics, generic icons
PRODUCT MODE App UI / dashboard / tool / admin / form-heavy
→ Function-first, density rewarded, no marketing voice
→ Display = same family as body (Geist/Inter/system), no serif headlines
→ Tabular numerics for data, mono for IDs/codes
→ Animation: only state-change feedback (loading/empty/error/success)
→ AI Tells to avoid: hero images, oversized H1, decorative shadowsCâu hỏi check: "Người dùng đến đây để CẢM NHẬN, hay để LÀM CÁI GÌ?" Cảm nhận → Brand. Làm việc → Product.
4. BIAS CORRECTION RULES (proactive — viết default tốt thay vì fix sau)
Rule 1: Deterministic Typography
- Display/Headlines:
text-4xl md:text-6xl tracking-tighter leading-none(brand mode);text-2xl font-semibold(product mode) - Body:
text-base text-neutral-600 leading-relaxed max-w-[65ch] - ANTI-SLOP fonts cấm dùng làm DISPLAY:
Inter,Roboto,Arial,Open Sans(body OK) - Brand mode display:
Geist,Outfit,Cabinet Grotesk,Satoshi,Newsreader,Tiempos,Iowan Old Style - Product mode:
Geist+Geist Mono, hoặcSatoshi+JetBrains Mono. Serif BANNED cho dashboard.
Rule 2: Color Calibration
- Max 1 accent color, saturation < 80%
- Neutral base: Zinc/Slate/Stone (KHÔNG pure black
#000, KHÔNG pure gray) - THE LILA BAN: No purple/violet glow, no neon gradient, no "AI Purple" aesthetic
- One palette per project. Không mix warm-gray + cool-gray cùng output
Rule 3: Layout Diversification
- Centered Hero CẤM khi
LAYOUT_VARIANCE > 4— force Split Screen / Asymmetric - 3-column equal cards CẤM — dùng Bento, Zig-Zag, asymmetric grid
- Mobile override: layouts ≥md PHẢI fallback strict 1-column ở
< 768px - Container:
max-w-[1400px] mx-autohoặcmax-w-7xl
Rule 4: Materiality
- Card containers chỉ dùng KHI elevation = hierarchy. Else dùng
border-t,divide-y, negative space - Shadow tinted theo background hue, KHÔNG default
shadow-md/lg/xl - Rounded radius: 8-12px crisp (product), hoặc 16-32px gentle (brand soft mode)
- KHÔNG
rounded-fullcho large containers - KHÔNG glassmorphism trừ khi có lý do — nếu dùng PHẢI có 1px inner border + inset shadow
Rule 5: Interactive UI States (mandatory generation)
LLM mặc định gen "static success" — em PHẢI gen full cycle:
- Loading: Skeletal loader match layout (KHÔNG generic spinner)
- Empty: Composed empty state với hint cách populate
- Error: Inline error reporting (forms)
- Tactile feedback:
:active→-translate-y-[1px]hoặcscale-[0.98]
Rule 6: Form Patterns
- Label TRÊN input (KHÔNG floating label, KHÔNG label trong placeholder)
- Helper text tồn tại trong markup (kể cả không hiển thị)
- Error text DƯỚI input
- Spacing:
gap-2cho input block
5. AI TELLS — FORBIDDEN PATTERNS (audit checklist trước ship)
Visual & CSS (cấm tuyệt đối trừ khi user explicit yêu cầu)
- ❌ Neon/outer glows (
box-shadowglow) - ❌ Pure black
#000000(dùng off-black#0a0a0a, Zinc-950) - ❌ Oversaturated accents (saturation > 80%)
- ❌ Excessive gradient text (text-fill gradients cho large headers)
- ❌ Custom mouse cursors
- ❌ Purple/violet gradient backgrounds (LILA BAN)
- ❌ Aggressive emoji feature icons (✨ 🚀 🎯)
- ❌ Rounded card với left coloured border accent
- ❌ Hand-drawn SVG humans/faces/scenery
- ❌ Italic-serif display headers
- ❌ Gradient on every background
Typography
- ❌ Inter/Roboto/Arial làm DISPLAY face
- ❌ Oversized H1 ngoài kiểm soát (control hierarchy bằng weight + color, không chỉ scale)
- ❌ Serif trên Dashboard/Software UI
Layout & Spacing
- ❌ 3-column equal cards generic
- ❌ Floating elements với gap awkward
- ❌ Padding/margin không mathematically perfect
- ❌ Centered Hero text trên dark image (try asymmetric thay vào)
Content & Data — "Jane Doe Effect"
- ❌ Tên placeholder "John Doe", "Sarah Chan", "Acme Corp"
- ❌ Generic SVG egg avatars / Lucide user icons
- ❌ Fake metrics "10× faster", "99.9% uptime" không có source
- ❌ Predictable numbers (
50%,99.99%,1234567) - ❌ Startup slop names "Acme", "Nexus", "SmartFlow"
- ❌ AI copywriting clichés: "Elevate", "Seamless", "Unleash", "Next-Gen", "Game-changer", "Delve"
- ❌ Lorem ipsum, "Feature One / Feature Two"
External Resources
- ❌ Unsplash links (broken thường) — dùng
https://picsum.photos/seed/{string}/800/600 - ❌ shadcn/ui generic default — PHẢI customize radii/colors/shadows
- ❌ Invented metrics → leave honest placeholder (
—, grey block, labelled stub)
Honest placeholder beats fake stat.
Full danh sách 27 anti-patterns: xem references/anti-slop.md.
6. PERFORMANCE GUARDRAILS
- DOM cost: Grain/noise filters CHỈ trên
fixed inset-0 pointer-events-nonepseudo-elements, KHÔNG trên scrolling containers - Hardware acceleration: Animate ONLY
transformvàopacity. KHÔNG animatetop/left/width/height - Z-index: KHÔNG spam
z-50/z-10. Dùng systemic layer contexts (sticky nav, modal, overlay) - Scroll listeners: KHÔNG dùng
window.addEventListener('scroll'). DùngIntersectionObserver - GSAP vs Framer Motion: KHÔNG mix trong cùng component tree. Default Framer Motion cho UI; GSAP/ThreeJS exclusively cho isolated full-page scrolltelling
7. DIAL CONFIG (3 knob để adjust output)
DESIGN_VARIANCE: 8 (1=symmetry, 10=chaos)
MOTION_INTENSITY: 6 (1=static, 10=cinematic)
VISUAL_DENSITY: 4 (1=art gallery, 10=cockpit)User có thể override bằng prompt: "make it more variance", "less motion", "denser data". Adapt config dynamically.
DESIGN_VARIANCE
- 1-3: flexbox center, symmetrical 12-col grid, equal padding
- 4-7: offset (margin-top: -2rem overlap), varied aspect ratios, left-aligned headers
- 8-10: masonry, fractional grid units, massive empty zones
- MOBILE OVERRIDE: levels 4-10 PHẢI strict 1-col fallback ở
< 768px
MOTION_INTENSITY
- 1-3: No automatic animation. CSS
:hover/:activeonly - 4-7:
transition: all 0.3s cubic-bezier(0.16, 1, 0.3, 1).animation-delaycascade. Transform/opacity only - 8-10: Scroll-triggered reveals, parallax. Framer Motion hooks. NEVER
window.addEventListener('scroll')
VISUAL_DENSITY
- 1-3: Art gallery — huge whitespace, expensive feel
- 4-7: Daily app — normal spacing
- 8-10: Cockpit — tiny padding, 1px lines (no cards), monospace numerics
8. WORKFLOW EXECUTION (recap với Kimi/Lovable-tier target)
TURN 1
├─ 1-line prose ("Got it — landing cho SaaS B2B. Tell me the rest:")
├─ <question-form id="discovery"> (≤7 questions)
└─ STOP — KHÔNG narrate
TURN 2 (branch on `brand` answer)
├─ A. "Pick direction" → emit <question-form id="direction"> (5 cards) → STOP
├─ B. "Have brand spec / Match reference" → extract → write DESIGN.md → goto Turn 3
└─ C. else → tone-based default direction → goto Turn 3
TURN 3+ — TARGET KIMI/LOVABLE-TIER, NOT "decent"
├─ READ references/quality-tiers.md → know the bar
├─ TodoWrite plan (5-10 imperative items)
├─ Mark steps in_progress / completed live (no batch)
├─ HERO: pick from references/hero-patterns.md (7 ready-to-paste)
├─ SECTIONS: pick from references/section-blueprints.md (10 ready-to-paste)
├─ MOTION: add từ references/motion-recipes.md (ambient pulse + entrance reveal mandatory)
├─ Wireframe pass first (grey blocks, labelled stubs) — show visible early
├─ Fill real copy + imagery (specific numbers, real names, picsum.photos)
├─ Step 7: 27-anti-pattern audit (P0 must all pass) — references/anti-slop.md
├─ Step 8: 5-dim critique (fix < 3/5) — references/critique.md
├─ Step 9: Quality tier scoring (≥ 24/26 = Kimi/Lovable-tier achieved) — references/quality-tiers.md
└─ Emit single <artifact>9. REFERENCE FILES — đọc on-demand
Foundation (luôn đọc khi bắt đầu task lớn)
| File | Khi nào đọc |
references/quality-tiers.md | BẮT BUỘC đọc ở Turn 3 — Kimi-tier vs Lovable-tier checklist + scoring rubric |
references/directions.md | User chọn "Pick direction" → bind palette/font verbatim |
references/discovery-form.md | Viết Discovery Form — full JSON schema, edge cases, examples |
Building (đọc khi viết section/hero)
| File | Khi nào đọc |
references/hero-patterns.md | READY-TO-PASTE 7 hero patterns (oversized display, bento, asymmetric split, cinematic, animated pulse, magazine drop cap, dual-tone) |
references/section-blueprints.md | READY-TO-PASTE 10 sections (bento, zig-zag, pricing, testimonial wall, FAQ, stat strip, CTA, footer, logo strip, comparison) + section combo recipes |
references/motion-recipes.md | 10 animation recipes (ambient pulse, IntersectionObserver reveal, sticky stack, magnetic button, shimmer, tactile press, counter-up, choreography, parallax, cursor spotlight) |
QA (đọc ở Step 7-8)
| File | Khi nào đọc |
references/anti-slop.md | Step 7 — full 27 anti-patterns với fix recommendations |
references/critique.md | Step 8 — 5-dim self-critique framework với scoring rubric |
Brand spec
| File | Khi nào đọc |
references/design-md-format.md | Branch B (extract brand-spec) hoặc tạo DESIGN.md mới — Google Stitch compatible |
Starter Design Systems (8 cái)
| File | Direction | Best for |
design-systems/kimi-style.md | Editorial Premium AI | AI tool landing, Kimi-tier target |
design-systems/lovable-style.md | Modern Minimal Cinematic | AI builder, dev tool, Lovable-tier target |
design-systems/linear.md | Modern Minimal Dark | Dev tool, productivity SaaS |
design-systems/stripe.md | Warm Soft | Fintech, wellness, indie SaaS |
design-systems/apple.md | Modern Minimal Premium | Premium product, brand site |
design-systems/monocle-editorial.md | Editorial | Long-form, content, journalism |
design-systems/brutalist-arena.md | Brutalist | Agency, manifesto, creative |
design-systems/datadog-utility.md | Tech Utility | Dashboard, admin, data-heavy |
10. FAQ
Q: Skill này thay thế shadcn/ui hay TailwindCSS? A: Không. Skill chỉ guide cách dùng chúng — bias correction + anti-slop rules. Code output vẫn là Tailwind + React (hoặc whatever stack user dùng).
Q: Có DESIGN.md sẵn rồi, vẫn cần Discovery Form? A: Có, nhưng tailor — drop questions DESIGN.md đã trả lời (palette, typography). Vẫn cần lock: output type, audience, tone, scale.
Q: User nói "just build, skip form" thì sao? A: Skip form, dùng tone-based default direction (modern-minimal cho dev-tools, warm-soft cho consumer/wellness, editorial cho content/media, tech-utility cho data/admin, brutalist cho creative/agency).
Q: Output stack?
A: Default React + TailwindCSS (v3 hoặc v4 — check package.json). Adapt nếu user dùng stack khác (Vue, Svelte, Astro, vanilla HTML).
Triết lý cuối: Mỗi rule trên đây là 1 lựa chọn cố ý chống lại default biases của LLM. AI mặc định ship "trung bình đẹp" vì training data đầy mediocre web. Skill này force ship "có taste" bằng deterministic guardrails + ready-to-paste blueprints + quality tier scoring.
11. QUALITY TIER QUICK REFERENCE
TIER 1 — "AI default" — Bootstrap/shadcn vibe, instantly recognizable as AI gen
TIER 2 — "Decent" — Pass review, không xấu nhưng không memorable
TIER 3 — "Kimi/Lovable-tier" — Senior designer level (TARGET)Kimi-tier signature checklist (editorial premium AI)
□ Page padding ≥ py-32, ambient gradient pulse, AI input prominent
□ Display clamp(48px, 7vw, 96px+), italic/serif emphasis, tracking-tighter
□ One accent ≤ 2 instances, real product UI screenshots
□ Mono numerics tabular, sticky frosted nav backdrop-blur-xl
□ NO Inter display, NO emoji icons, NO 3-col equal feature gridLovable-tier signature checklist (modern minimal cinematic)
□ Bento grid hero (6-9 tiles) HOẶC oversized type ≥ text-7xl
□ Dark mode native, animated gradient pulse 2+ layers
□ Big bold display weight 600+, gradient text emphasis ≤ 1 element
□ Cards rounded-3xl border-white/10, conversational hero copy
□ Specific stats organic (847k, 14ms, 99.97%), real product preview tilesScore self ≥ 24/26 trên 13 aspects (xem references/quality-tiers.md cho full table) → Kimi/Lovable-tier achieved.
< 20/26 → AI default — major rework needed.
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.

