design-led-website-builder
Skill251 starsChanged 3 months agoApache-2.0View on GitHub
Turn a thin brief (who, why, what to show) into a distinctive researched website. If the brief could mean more than one product, ask once to lock intent and wait. Then research matching live sites, hunt licensed or generated materials (textures, open art, editorial photos — not fake portraits), and build. Skip the ask when the job, a style, a URL/screenshot, or a brand already settles it. Use for new sites and redesigns, not bug fixes or copy edits.
In this repository5 other files it gives its agents.
- gzh-title-strategistSkill
- illustrated-manual-wordSkill
- liucong-model-evalSkill
- llm-wiki-opsSkill
- painterly-3d2-cinemaSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
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
- Very specific: line 19
Lock intent before any research: If the brief can mean more than one product, send one ask_user round and stop until they reply.
- Distinctive: line 26
Do not substitute a schedule, dossier, or interview for the website.
- Distinctive: line 32
If you already asked once, the next move is research or code.
- Distinctive: line 81
Ask once, then stop until they answer.
- Distinctive: line 84
Do not copy signature layouts, type, copy, or assets.
- Distinctive: line 85
Do not invent customers, metrics, awards, portraits, or a career.
- Distinctive: line 87
Do not overwrite unrelated user work.
≈ “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
Design Led Website Builder
A short brief is enough. Lock intent, wait, research that job, hunt materials, then build. Ask at most once.
Typical input: who they are, why the site exists, what it must show. Also accept a brand kit, a named style, a reference URL, or a screenshot.
This document is English. Talk to the user in their language. Ask_user options follow the user's language.
People arrive differently. Some bring a brand. Some bring a name and a few pieces of content. Same pipeline; different raw material. Read intake.
Pipeline (do not reorder)
- Ingest the brief and notice the raw material (brand / no brand / a reference).
- Lock intent before any research. If the brief can mean more than one product, send one
ask_userround and stop until they reply. If intent is already singular, skip. - Research live sites that match the locked intent. Read research pass.
- Hunt materials in the same research turn. Read visual evidence and assets. Find licensed or generated textures, open art, and editorial images. Do not ship a type-only page.
- Return 8–12 lines: direction, three site references, materials sourced, what you will not copy, what you will build.
- Build in the same turn. Read brief to direction, visual craft, design-to-build.
- Verify with the quality rubric.
Do not substitute a schedule, dossier, or interview for the website. Do not keep operating and asking at the same time.
Stop conditions (hard)
At most one question round in the whole job (plus a later rights question only if you are about to publish someone else’s photo or a real price).
If you already asked once, the next move is research or code.
1. Ingest — two kinds of raw material
| They brought | What you do |
| A brand — name, logo, colors, existing site, guidelines | Keep it. After intent is locked, research peers in that category. Do not invent a second identity. |
| No brand — a person or project, some content, maybe a taste cue | After intent is locked, invent a distinctive system from purpose + content + cue. Do not stall for a brand kit. |
Also capture why, what to show, and any already-given lock (style, URL, screenshot).
2. Lock intent (before research)
List the 2–4 products this brief could be. If more than one is plausible, ask once. Options come from this brief, not a fixed menu.
Typical forks (examples, not a checklist to dump):
| They said | Could mean |
| blog / personal site | Publishing site (write, read, archive) or visually driven personal site |
| company site / homepage | Product site, company brochure, recruiting, or brand film |
| portfolio | Case-study studio, image index, or spatial/motion piece |
| landing page | Waitlist, sales, or story |
| premium / beautiful / more designed | Still need the job: what is the site for? |
Each option is a product with a consequence, plus I have a reference (next message: URL or screenshot).
Skip the ask when intent is already singular. After sending the question: wait. Then research that job only.
3. Research sites, then hunt materials
Open three to five live sites that match the locked intent. User-supplied URL or screenshot is reference #1.
Then collect 3–8 production materials (paper, wood, texture, licensed still life, original illustration). Proof of the named person still waits on their files. Atmosphere does not. CSS rings and empty slots are not materials.
uv run --with pyyaml --python 3.11 python scripts/catalog.py match \
--scenario editorial --style editorial --component galleryVisual evidence: protocol. Audit example: Gionatan Nese. Worked path: thin brief.
4. Return, then build
Name the direction, the site references, and the materials you put on disk. Then implement. Two different briefs must not share type, color, and layout.
Hard rules
- Intent is locked before research. Research follows the lock.
- Ask once, then stop until they answer. Do not research in parallel with the question.
- Hunt materials. Type-only first builds fail.
- User reference and brought brand beat catalog guesses.
- Do not copy signature layouts, type, copy, or assets.
- Do not invent customers, metrics, awards, portraits, or a career.
- Missing proof (their photo, their work): high-fidelity labeled slots. Missing materials: you failed the hunt.
- Do not overwrite unrelated user work.
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.

