agentleFS
delphinelil09-sysdelphinelil09-sys/universal-design/design-core/SKILL.md

design-core

Skill3 starsChanged 49 days agoMITView on GitHub

Universal method for designing, auditing, and elevating websites. Use whenever the user works on a site — landing page, portfolio, ecommerce, info site, SaaS — mentions redesign, 'make it look better', 'why does my site look generic', 'audit my site', 'why isn't this converting', or asks for a site that doesn't look AI-generated. Routes to one of five subtype skills (design-promo, design-portfolio, design-ecommerce, design-info, design-saas) after establishing why the site exists. Covers two modes (audit / creation) and two phases (floor / ceiling).

2 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions2 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 2

    1. ldaume/agentic-engineering-harness★ 3same name
    2. th3r-dev/th3-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 repository5 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 9 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    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
  1. Distinctive:

    Before touching design, walk these three steps in order.

    line 14
  2. Distinctive:

    Do not start design work until you know why the site exists.

    line 26
  3. Very specific:

    Audit: does not follow the layer order.

    line 65
  4. Distinctive:

    Read the request before choosing scope.

    line 71
  5. Distinctive:

    Every task has a floor and a ceiling.

    line 83
  6. Distinctive:

    When Phase 1 is done, the site is correct.

    line 98
  7. Distinctive:

    Do not invent a concept. Phase 2 requires the user to provide:

    line 102
  8. Distinctive:

    If the site is a hybrid, ask first, then load the leading subtype.

    line 134
  9. Distinctive:

    No stated goal: Do not guess. Ask "why does this exist?" and offer a hypothesis ("looks like a landing page for sale — right?") to make it …

    line 138
  10. Distinctive:

    Ambiguous subtype: Ask: "impression or sale?"

    line 140

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Design Core

The universal method for working on websites. This skill is the entry point. It establishes context, asks the right question, and routes to a subtype lens. It does not design anything by itself — it makes sure the work starts in the right place.

Scope: websites only. Not business cards, posters, decks, apps, or dashboards.

Three entry steps

Before touching design, walk these three steps in order. They are the same whether the task is audit or creation.

1. Observation

Establish three facts. If any is unclear, ask.

  • Artifact — is there a ready HTML, a screenshot, a URL, or only a brief?
  • Subtype — is this promo, portfolio, ecommerce, info, or saas (subscription software)? (See routing below.)
  • Task — is this create, fix, or elevate?

2. Why

Do not start design work until you know why the site exists.

  • What is it for — sale, impression, explanation, self-declaration?
  • Who is the audience?
  • What should happen after someone views it?

The answer decides who leads:

GoalWho leads
Sale, signup, conversionMarketing leads
Impression, character, workDesign leads
Explanation, trust, structureThey cooperate — design serves clarity

3. Routing

Load the subtype lens that matches the site. If the subtype is not obvious (a hybrid — e.g., a studio site that both shows work and sells services), ask the user: "Is this more about impression or about sale?" Then load the leading subtype, and pull in the second only for the sections that need it.

The running context

From entry to done, keep these five facts in mind. They are the context the lens builds on — a lens never restarts the method, it adds to it:

  1. Artifact — HTML, screenshot, URL, or brief?
  2. Subtype — which lens is leading?
  3. Who leads — marketing (sale), design (impression), or cooperation (explanation)?
  4. Mode — audit or creation?
  5. Phase — floor (fixed), ceiling (user's direction), or both?

State them aloud when the lens loads, so the user can correct a wrong assumption early. If a later decision contradicts one of these five, the context wins and the decision is re-made.

Two modes

The entry step decides which tool runs.

ModeWhenToolPhase weighting
AuditUser hands you an artifactGrid — read all eight elements at once, hunt for conflictsPhase 1 = main work, Phase 2 = optional
CreationUser hands you a briefLayers — move top-downPhase 1 = quick check, Phase 2 = main work

Audit does not follow the layer order. It reads the artifact across all eight elements simultaneously, finds what's broken, and fixes it. See references/audit-grid.md.

Creation moves through layers from strongest to weakest: frame → body → surface → accent. See references/layers.md and references/generation.md.

Scope discipline

Read the request before choosing scope.

  • Vague request — "look at my site", "review this", "what do you think", "any suggestions". Run the full floor audit first: all eight elements, floor findings, format Element → Layer → Problem → Fix. Then offer Phase 2 separately. Do not jump straight to character suggestions.
  • Specific request — "check the signup form", "what's wrong with the hero", "fix the contrast". Do exactly that. Do not expand scope. Do not run a full audit unless the user asks for one.

Examples:

  • "Check out my website" → full audit, all eight elements, floor first.
  • "What's wrong with hero?" → hero only. Do not audit other sections.
  • "Check the contrast." → contrast check only. Do not expand to other Color findings.

Two phases

Every task has a floor and a ceiling.

Phase 1 — Floor (autonomous)

Fix what is objectively broken. Do not ask permission — fix it. What counts as broken:

  • Contrast — text against background below 4.5:1 (3:1 for large headings).
  • Click zones — interactive targets smaller than 44×44px on mobile.
  • Body text size — below 16px on mobile.
  • Focus — no visible focus indicator on interactive elements (not only hover).
  • Spacing scale — random values instead of 4 / 8 / 12 / 16 / 24 / 32 / 48 / 64.
  • CTA hierarchy — no clear primary / secondary / tertiary distinction.
  • Alignment — mixed systems with no reason; elements that "walk".
  • Fill — empty space that reads as a hole, not as air.

When Phase 1 is done, the site is correct. Not beautiful — correct. The user may stop here, and that is a complete outcome.

Phase 2 — Ceiling (never autonomous)

Do not invent a concept. Phase 2 requires the user to provide:

  1. An idea — what should this express, what character.
  2. A direction — a reference, an example, "like X but Y".
  3. Material — photos, textures, brand elements, if any.

With those, expand the idea into concrete decisions using the protocol in references/ceiling.md: decompose the reference by elements, extract the gesture (one sentence), translate top-down through the layers. Without them, Phase 2 does not run. Say so plainly: "Character cannot be chosen for you — give me a direction, or the site stays at the floor."

Phase 2 never runs on a site that failed Phase 1. Correctness first, character second.

References

References are not pasted into this file. To load one, read its file with your available read/file tool — never guess its content. Load on demand; do not read them all at once.

  • references/elements.md — the eight elements as vocabulary. Load when first speaking about design vocabulary.
  • references/layers.md — top-down movement protocol. Load in creation mode.
  • references/audit-grid.md — diagnostic grid. Load in audit mode.
  • references/generation.md — the "0 to 1" process. Load in creation mode.
  • references/ceiling.md — Phase 2 protocol: idea + reference → concrete decisions. Load in creation mode when the user brings a direction.
  • references/marketing.md — marketing as a filter layer. Load when the goal is sale.
  • references/copy.md — words as a design element; the textual slop filter. Load in audit and creation whenever text is written or assessed.
  • references/trends-2026-Q3.md — dated snapshot of current patterns. Load when the user asks for "current" or "modern"; check freshness.
  • references/tools.md — file editing rules (CRLF, batch limits, scope discipline for edits). Load when the task involves writing or modifying files.

Subtype routing

  • promo (landing pages, product launches, service promos) → load design-promo.
  • portfolio (designer, photographer, agency, artist) → load design-portfolio.
  • ecommerce (stores, product pages, catalogues) → load design-ecommerce.
  • info (corporate, documentation, content-heavy) → load design-info.
  • saas (subscription software, pricing, product-led growth) → load design-saas. For subscription products this replaces promo — the visitor is buying a promise, not an event; pricing and trust carry the page.

If the site is a hybrid, ask first, then load the leading subtype.

Error handling

  1. No stated goal. Do not guess. Ask "why does this exist?" and offer a hypothesis ("looks like a landing page for sale — right?") to make it easy to confirm or correct.
  2. Ambiguous subtype. Ask: "impression or sale?"
  3. "Make it beautiful" with no reference. Phase 2 does not run. Explain why and ask for a direction.
  4. HTML too large. Work by sections — first screen, navigation, cards — or ask for the key fragments.
  5. User asks to break the floor. Aesthetic rules (spacing, alignment, rhythm) may be broken at the user's will. Accessibility rules (contrast, click zones, focus) may not. The floor exists for people, not taste.
  6. User doesn't know a term. Explain using their own site: "the frame of your site is this block — and it's broken here."

What this skill is not

It is not a checklist. It is not a style. It is not a set of patterns to apply. It is a method: establish why, choose the mode, move through the layers or the grid, and stop at the floor unless the user brings an idea.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.