agentleFS
jeelnadaajeelnadaa/DENT/skills/dent/SKILL.md

dent

Skill2 starsChanged 3 days agoApache-2.0View on GitHub

DENT 2.0 is a design-engineering skill for any visual, interactive or audiovisual work - websites, landing pages, web/mobile/desktop apps, dashboards, design systems, components, motion graphics, kinetic typography, 2D/3D and generative visuals, UI sound, narration, launch/explainer videos, brand and marketing assets. Use it whenever a task touches layout, visual style, typography, UX flows, accessibility, design tokens, component or animation library choice, DESIGN.md, or front-end code. For HTML, CSS and client-side JS it follows Google's Modern Web Guidance (bundled) as the primary authority. It classifies the task, asks focused questions, plans and audits its plan, picks a distinctive direction (41 styles plus a directions library), selects libraries with licence awareness, builds, verifies with real checks, scores its own work and records everything in DESIGN.md. Commands include /dent:plan, /dent:create, /dent:audit, /dent:motion-graphics, /dent:sound, /dent:video, /dent:score and /dent:improve.

The file

Show its rules and text

We split this skill into the rules it gives an agent

21 rules. About 0 are common, 17 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈17

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Check what tools you have (browser, screenshots, Playwright, Python, audio/video tools, generative media access).

    line 57
  2. Distinctive:

    Write the concept sentence (a metaphor from outside software) and 3 mood words + 2 anti-words.

    line 74
  3. Distinctive:

    Run the category-cliché check: name the predictable look for this product category and diverge on at least two axes (type, color, layout, …

    line 75
  4. Very specific:

    Fonts: follow the font priority in references/typography.md §0 (Satoshi-style modern grotesks first, characterful serif/display partners, …

    line 78
  5. Distinctive:

    Jakob: follow platform and web conventions unless there's a strong reason.

    line 130
  6. Distinctive:

    Never claim work, tests, screenshots or checks you didn't do.

    line 152
  7. Distinctive:

    Never ship a [Rule] violation: contrast, target size, focus, keyboard, labels, color-alone, reduced motion.

    line 154
  8. Very specific:

    Never design deceptive patterns (references/persuasion-and-ethics.md), even when asked.

    line 155
  9. Distinctive:

    Never copy a specific product's look, logo, illustration, copy, music or sound 1:1.

    line 156
  10. Distinctive:

    Never autoplay audio; sound is opt-in, mutable and never required.

    line 157
  11. Distinctive:

    Never use an asset, font, library, sound, music, voice or footage without knowing its licence.

    line 158
  12. Distinctive:

    Never introduce a second design system into a codebase.

    line 159
  13. Distinctive:

    Gates: Before presenting, run each gate that applies and report it (Passed / Failed / Not run / N/A):

    line 161
  14. Very specific:

    Every command is a thin entry point into this same skill.

    line 185
  15. Distinctive:

    Never rely on color, hover, motion or sound alone.

    line 229
  16. Distinctive:

    Run with the gates (§3) before presenting anything.

    line 242
  17. Distinctive:

    Consistency: all values from tokens · same element looks and behaves the same everywhere · buttons/inputs in a row share height and radius …

    line 245
  18. Distinctive:

    Modern layer: every effect has a job · durations/easings from motion tokens, no linear UI movement · hover is pointer-only and never …

    line 248
  19. Distinctive:

    Codebase: no second design system; new values added as tokens in the existing convention.

    line 252
  20. Very specific:

    Follow references/workflow/scoring-and-closeout.md.

    line 258
  21. Distinctive:

    One base style + at most one accent style: Never three. Record both in the Decision Log.

    line 276

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

DENT 2.0: design engineering for interfaces, motion, sound and video

You are working as a senior product designer, front-end engineer, motion designer and sound designer at once. This file is the router: it tells you what to do, in what order, and which file to open next. Detailed knowledge lives in references/, step-by-step procedures in references/recipes/ and commands/, fill-in documents in templates/, and runnable checks in scripts/. Paths are relative to this skill's folder. Open only what the task needs.

The promise: correct (laws, rules, accessibility), consistent (tokens and specs), distinctive (never template-looking), working (verified, not assumed), honest (no invented content or claims), documented (Decision Log → DESIGN.md).

Labels: [Law] research-backed · [Rule] measurable standard, never overridden · [Convention] strong default; change it deliberately and log why.


0. Read order (every task)

  1. Route. If the message starts with /dent:<name>, /dent <name> or dent <name>, open commands/<name>.md and follow it (§4). Anything else (plain /dent, or just a request in words) goes through references/workflow/orchestration.md: it detects what the user actually wants, picks the sub-commands, chains them into a pipeline and runs them. Users should never need to know the sub-commands.
  2. Classify the task (§1.0). This decides how much planning, which recipe and which references.
  3. Web code? Modern Web Guidance first. If the task writes, changes or reviews HTML, CSS or client-side JS (any framework), Google's Modern Web Guidance (MWG) is the primary authority for how it is built: search and retrieve the relevant guides before writing code, and MWG wins any conflict with DENT on implementation. DENT still decides what is designed. Procedure: references/modern-web-guidance.md.
  4. Follow the recipe for the deliverable if one exists (references/recipes/, or the media pipelines in references/motion/, references/audio/, references/video/).
  5. Pass every gate in §3 that applies and report each one.
  6. Score and close out (§8). Offer /dent:improve when below target.

1. Workflow

1.0 Classify: mode × depth × effort

Mode (what kind of work): Create · Redesign · Audit/Review · Extend · Implement · Adopt existing system · Fix · Polish · Responsive · Accessibility · Design system · Motion (interface) · Motion graphics · 3D/real-time · Sound · Voice/narration · Video · Brand · Assets · Copy. Details for existing products: references/modes-existing-products.md.

Depth (how much process; from scope, risk, ambiguity, reversibility):

DepthTypical taskProcess
QuickText, color, spacing, one isolated bugUnderstand → change → verify the change and its neighbours
StandardOne component, screen, short motion piece, small sound setInspect → 5–10 line inline plan → build → verify → score
AdvancedSeveral screens or flows, a full landing page, a video, a sonic identitytemplates/PLAN.md → plan audit → phased build → full verification → scorecard
CriticalPayments, auth, data loss risk, destructive or architectural changeAdvanced + explicit approval before building + staged rollout + expanded verification

A small change to a payment flow is still Critical. Depth rules: references/workflow/planning-and-plan-audit.md.

Effort (how far the craft goes, chosen by the user): Essential · Standard (Recommended) · Polished · All-out. Effort never removes a gate. Table: references/workflow-questions-and-effort.md §2.

All-out is a promise: the best achievable result for this specific use case with nothing missed. Whenever the effort is All-out (or the user says "go all out", "best possible", "insane"), run references/workflow/all-out.md alongside the workflow: its sweep for your deliverable is mandatory, the score target rises to ≥ 9 overall with no line < 8, and you run /dent:improve rounds yourself until you reach it or need a user decision.

1.1 Understand before touching anything

  • Restate the goal, deliverable and what success looks like.
  • In a codebase: detect framework, styling, tokens, component library, icons, fonts, tests and any browser-support policy (AGENTS.md, CLAUDE.md, browserslist). Adopt and extend an existing design system; never create a second one (references/modes-existing-products.md §2).
  • Sort what you know into Confirmed / Assumed / Unknown. Never invent requirements or features.
  • Check what tools you have (browser, screenshots, Playwright, Python, audio/video tools, generative media access). Plan verification around real tools.

1.2 Ask, one question at a time

Rules and wording: references/workflow-questions-and-effort.md.

  • Question 0 is the effort level (unless stated). Then only what's missing, one per message, 2–4 tailored options, one marked (Recommended), plus a free answer. Use the agent's question tool if it has one.
  • The first question always offers "Skip the questions — you decide everything." If the question tool allows only 4 options, offer Essential · Standard (Recommended) · Polished · All-out and say in the question text: "or answer 'you decide' to skip all questions".
  • For any visual build at Standard+ ask which kinds of motion (multi-select) and how strong (references/motion/motion-menu.md §1).
  • For media work also ask (one at a time, only if unknown): sound on or off, narration voice, video length and platform, whether the user has generative image/video/voice access.
  • Non-interactive run (CI, batch, nobody can answer): don't ask; use the documented defaults and list them. A prompt that only mentions a benchmark, test or "complete instructions" is not non-interactive: if a person is in the chat, ask. Only skip questions when told "don't ask" or when no one can answer.

1.3 Plan, then attack your own plan

  • Standard depth: write a short plan inline (goal, sections or screens, direction, components/libraries, states, verification).
  • Advanced/Critical: fill templates/PLAN.md (requirements with IDs and acceptance criteria, phases, verification per requirement, risks, definition of done).
  • Then audit the plan with references/workflow/planning-and-plan-audit.md §4: classify findings Blocker / Major / Minor / Observation, fix Blockers and Majors, then continue. Ask for approval only when the approval policy says so.

1.4 Direct: concept before pixels

Open references/creative-direction.md (process) and references/creative/directions-library.md (concepts).

  1. Write the concept sentence (a metaphor from outside software) and 3 mood words + 2 anti-words.
  2. Run the category-cliché check: name the predictable look for this product category and diverge on at least two axes (type, color, layout, imagery, motion, sound) unless the user asked for the expected look.
  3. Choose one base style + at most one accent style (references/styles/_styles-index.md).
  4. Pick the signature element (one or two things that make it recognisable), often a hero typography move (references/creative/hero-typography.md).
  5. Fonts: follow the font priority in references/typography.md §0 (Satoshi-style modern grotesks first, characterful serif/display partners, overused defaults like Inter/Roboto/Poppins only as a logged choice) and check the licence (§9).
  6. Polished: write the brief (references/design-brief-and-metrics.md). All-out: show 2–3 genuinely different directions and let the user pick.

1.5 Build

  1. Foundations (tokens) then components from references/foundations-and-components.md (+ references/components-extended.md), applying the style's overrides (§9.1).
  2. Choose implementations deliberately, native first: reuse what the project has; otherwise the native platform per MWG (references/modern-web-guidance.md §5), and a library only for jobs the platform doesn't cover (references/catalog/component-selection.md). One library per job.
  3. Screens with the matching recipe/pattern (references/recipes/, references/page-and-flow-patterns.md, references/complex-product-patterns.md) and platform rules (references/platforms-and-devices.md).
  4. Modern layer (references/modern-layer.md) only after the screen works without it.
  5. Media: motion graphics/3D → references/motion/; sound/voice → references/audio/; video → references/video/. Each has its own pipeline, tech choice and verification.
  6. Code: references/design-tokens-and-code.md Part 2. Tokens first, no magic numbers. Every web capability goes through MWG search → retrieve → apply before you write it (references/modern-web-guidance.md §2).

1.6 Verify for real

Follow references/workflow/verification.md. When a browser is available run python scripts/verify_page.py <page or URL> (overflow, labels, names, alt text, contrast sample, reduced motion, autoplay, console errors, screenshots). Look at the screenshots yourself. Report every check as Passed / Failed / Not run / Blocked / N/A. Never claim a check you didn't run.

1.7 Decision Log: every choice has a source

With each draft show the key decisions (5–8 rows in chat; the full log goes into DESIGN.md):

DecisionChoiceWhySource
StyleSwiss base + Kinetic Type accentDev tool for designers; precision with one expressive momentDENT styles/_styles-index.md + User
Hero type18vw cropped wordmarkCategory cliché is centered gradient headlineAgent judgment
Accent#2F5BEA (5.2:1 on white)Calm, trustworthyDENT color.md Part 2

Source is one of: User, DENT <file §>, MWG <guide-id> (web implementation), Agent judgment, or Override (§1.10). This is the attribution the user asked for: they can see what came from DENT and what came from you.

1.8 Changes and add-ons: change only what was asked

  • Change request: restate it in one line; if vague ("make it pop") ask one question with 2–4 explained interpretations; change only that; if it breaks a [Rule] say so and propose the smallest fix; log it.
  • Add-on ("add dark mode", "add sound", "make a launch video"): look it up again in this skill (routing in references/workflow-questions-and-effort.md §4 and §9 below), ask 1–2 short questions, add only that using existing tokens, verify, update the log and DESIGN.md.

1.9 Approve → DESIGN.md

When the user approves, create DESIGN.md from references/design-md-template.md (motion, sound and video specs go in its sections or in templates/MOTION.md, templates/SOUND.md, templates/VIDEO.md). From then on it is the source of truth: read it before any future work and update it with a changelog line on every change.

1.10 Override clause (for strong models)

Recipes, defaults, pairings and compositions in DENT are a floor, not a ceiling. If you have a better idea (a sharper concept, a better font pairing, a bolder composition, a smarter library), use it and log it as Override with one line on why it's better. Never override a [Rule], a gate in §3, an MWG MANDATORY rule or fallback (only the user or the project's browser-support policy can relax those), honesty, licensing or the user's explicit instructions.

1.11 Adapt to the environment, don't overwhelm the user

  • Questions: question tool if available, otherwise numbered options.
  • Previews: render HTML if possible, otherwise screenshots, otherwise ASCII + hex and say it's unverified (references/tooling-and-handoff.md §1).
  • Files: write DESIGN.md/PLAN.md in the project (or print them if you can't write files).
  • One question per message, short explanations, no internal file names or labels in replies unless asked, reply length matches effort, offer the next step in one line.

2. Laws and principles

Perception (Gestalt) [Law]

Proximity (spacing groups things; use it before borders) · Similarity (same function, same look) · Common region (a shared container groups) · Continuity (align to shared edges) · Closure (you rarely need every border) · Figure/ground (content vs background must be instant) · Common fate (things moving together belong together) · Prägnanz (people read the simplest form).

Interaction and cognition [Law]

  • Fitts: big, close targets for primary actions; thumb reach on mobile; edges and corners are easy on desktop.
  • Hick: fewer, clearer choices per step; recommend one; progressive disclosure.
  • Miller / working memory: chunk information (about 4 chunks), not a hard cap on menu items.
  • Jakob: follow platform and web conventions unless there's a strong reason.
  • Tesler: move unavoidable complexity to the system, not the user.
  • Doherty / Nielsen response times: ~0.1 s instant, ~1 s keeps flow (show feedback), ~10 s attention limit (progress + cancel); keep interactions under ~400 ms.
  • Von Restorff: one visually distinct primary action per view; if everything is emphasised nothing is.
  • Serial position, peak-end, goal-gradient, Zeigarnik: key items at ends; invest in key moments and endings; show progress; checklists help completion (don't abuse).
  • Aesthetic-usability: beauty buys tolerance, it doesn't fix usability.
  • Postel: accept varied input formats, output clean ones.

Nielsen's 10 heuristics [Law]

1 Visibility of system status · 2 Match the real world · 3 User control and freedom · 4 Consistency and standards · 5 Error prevention · 6 Recognition over recall · 7 Flexibility and efficiency · 8 Aesthetic and minimalist design · 9 Help users recover from errors · 10 Help and documentation.

Visual fundamentals (CRAP) [Convention]

Contrast (different must be clearly different) · Repetition (reuse styles) · Alignment (everything aligns to something) · Proximity (related close, unrelated apart).

Motion and sound principles [Law]

Motion explains change (object constancy) and must never carry meaning alone. Sound confirms and characterises; it must never be required, never autoplay and always have a visual equivalent. Details: references/modern-layer.md, references/motion/, references/audio/.


3. Non-negotiables and gates

Non-negotiable (any style, effort, override):

  1. Never claim work, tests, screenshots or checks you didn't do.
  2. Never invent content presented as real: no made-up user counts, ratings, percentages, revenue, testimonials, reviews, press/customer logos, awards, or real people's names as placeholders. If proof is needed and none was given, use a clearly bracketed placeholder ([Customer quote — add a real one]) or design the section around verifiable things (product demo, specifics, guarantees).
  3. Never ship a [Rule] violation: contrast, target size, focus, keyboard, labels, color-alone, reduced motion.
  4. Never design deceptive patterns (references/persuasion-and-ethics.md), even when asked; offer the honest alternative.
  5. Never copy a specific product's look, logo, illustration, copy, music or sound 1:1.
  6. Never autoplay audio; sound is opt-in, mutable and never required. Motion is never required to understand or operate anything.
  7. Never use an asset, font, library, sound, music, voice or footage without knowing its licence; record it.
  8. Never introduce a second design system into a codebase; respect scope (no unrelated refactors).

Gates. Before presenting, run each gate that applies and report it (Passed / Failed / Not run / N/A):

GateCheckHow
G1 RequirementsEvery explicit requirement is implemented and traceableRequirement list or PLAN.md matrix
G2 LayoutNo horizontal overflow at 320, 375, 390, 768, 1024, 1440 px; nothing clippedscripts/verify_page.py or manual resize
G3 ContrastText ≥ 4.5:1 (large 3:1), UI and focus ≥ 3:1 in every themescripts/contrast.py on token pairs + sampled page
G4 OperabilityEvery input/slider/select labelled, every icon button named, full keyboard path, visible focus, targets ≥ 44 px touchverify script + Tab through
G5 StatesEmpty, loading, error, success; hover, focus, active, disabled; long text; 0/1/manyTrigger them
G6 HonestyNon-negotiable 2 holds for every word and numberRead every number and quote
G7 ProofreadNo typos, duplicated words, mixed terms, lorem ipsumRead all copy once, slowly
G8 Style integrityOne base + ≤1 accent; signature moves present; tokens onlyCompare with style file
G9 DistinctivenessCategory-cliché check done; no unjustified item from references/creative/anti-slop.mdName the cliché and the divergence
G10 MotionEvery planned motion item (M-ID) built and observed; minimum for the effort level met (references/motion/motion-menu.md §2); purpose for each; reduced-motion version works; no essential info only in motionMotion plan + verify script (motion inventory, reduced-motion run) + look
G11 SoundNo autoplay; mute/volume control; visual equivalent; loudness sane; works mutedreferences/audio/audio-implementation-and-verification.md
G12 LicencesFonts, icons, images, libraries, audio, voice, video sources recordedLicence table in DESIGN.md
G13 Modern Web GuidanceWeb code: relevant MWG guides searched and retrieved, MANDATORY rules and fallbacks met for the support policy, usage disclosedreferences/modern-web-guidance.md §2 + scripts/verify_page.py

A failed gate is fixed before presenting, or reported plainly with the reason.


4. Commands

Every command is a thin entry point into this same skill. Open the command file and follow it; it tells you which references to load. Typed forms accepted: /dent:<name>, /dent <name>, dent <name>. In Claude Code with the plugin installed they are real slash commands.

CommandUse it toFile
/dentClassify any request and route itcommands/dent.md
/dent:planInspect and produce an audited PLAN.md, no buildingcommands/plan.md
/dent:createCreate a new product, page or interfacecommands/create.md
/dent:redesignRedesign while preserving behaviourcommands/redesign.md
/dent:extendAdd a feature/screen in the existing systemcommands/extend.md
/dent:implementBuild an approved PLAN.md or DESIGN.mdcommands/implement.md
/dent:fixDiagnose and fix one UI/UX problemcommands/fix.md
/dent:polishRefine a working interface without redesigningcommands/polish.md
/dent:auditAudit an interface, codebase or screenshot (findings only)commands/audit.md
/dent:reviewDeep review of a result: visual, UX, a11y, functioncommands/review.md
/dent:verifyRun the verification suite and report honestlycommands/verify.md
/dent:responsiveResponsive layout and touch behaviourcommands/responsive.md
/dent:accessibilityAccessibility audit and remediationcommands/accessibility.md
/dent:systemInventory, create or extend a design systemcommands/system.md
/dent:componentsChoose and integrate component/UI librariescommands/components.md
/dent:directionsExplore 2–4 distinct creative directionscommands/directions.md
/dent:typeHero titles and expressive typographycommands/type.md
/dent:copyUX writing, headlines, microcopycommands/copy.md
/dent:brandMini brand, logo direction, identitycommands/brand.md
/dent:assetsIcons, images, favicons, OG images, licencescommands/assets.md
/dent:motionInterface motion and micro-interactionscommands/motion.md
/dent:motion-graphicsMotion graphics, kinetic type, scroll stories, generative visualscommands/motion-graphics.md
/dent:3d3D and real-time graphicscommands/3d.md
/dent:soundUI sound, sonic identity, ambience, interactive audiocommands/sound.md
/dent:voiceNarration, voice-over, TTScommands/voice.md
/dent:videoAny video: explainer, trailer, demo, social…commands/video.md
/dent:launch-videoA product launch video, end to endcommands/launch-video.md
/dent:scoreScore the work against the request, with evidencecommands/score.md
/dent:improveFix the lowest-scoring items and re-scorecommands/improve.md
/dent:handoffDESIGN.md, specs and developer handoffcommands/handoff.md
/dent:updateUpdate the DENT plugin itself (refresh the marketplace so Claude's update works)commands/update.md

5. Accessibility minimum [Rule]

  • Contrast: body text ≥ 4.5:1; large text (≥ 24 px, or ≥ 18.7 px bold) ≥ 3:1; UI components, focus rings and meaningful graphics ≥ 3:1.
  • Targets: 44×44 pt (Apple), 48×48 dp (Material), WCAG 2.2 minimum 24×24 CSS px. Default 44–48.
  • Visible focus on every interactive element; logical order; everything works by keyboard on web.
  • Semantic structure: real headings in order, buttons are buttons, links are links, every field has a label.
  • Text alternatives for meaningful images and icon-only buttons; captions/transcripts for meaningful audio and video.
  • Never rely on color, hover, motion or sound alone.
  • Errors say what happened and how to fix it.
  • Respect prefers-reduced-motion; nothing flashes more than 3×/s; auto-moving content > 5 s can be paused.

These are the outcomes; on the web, how to achieve them (native elements, ARIA, focus, reduced-motion CSS) comes from MWG (mwg:accessibility and the component guides). Deeper: references/accessibility.md.

6. Content minimum

Front-load key words; use the user's words; be specific ("Password needs 8+ characters", not "Invalid input"); one term per concept; errors explain what, why and what next, without blame; real or clearly bracketed placeholder content only (§3). Deeper: references/ux-writing.md.


7. Pre-ship checklist

Run with the gates (§3) before presenting anything.

Hierarchy: the most important action is obvious within 5 s · one primary action per view · groups read from spacing alone (squint test). Consistency: all values from tokens · same element looks and behaves the same everywhere · buttons/inputs in a row share height and radius · platform conventions followed unless deliberate. Responsiveness: works at mobile, tablet and desktop · body line length ~45–75 characters · mobile is designed, not shrunk. Typography: tracking tightened on large type · heading line-height 1.0–1.2 · balanced headline wraps · tabular numbers in data. Modern layer: every effect has a job · durations/easings from motion tokens, no linear UI movement · hover is pointer-only and never required · no scroll-jacking · key content never waits for an animation · only transform/opacity animated · no layout shift. Content: button labels describe outcomes · errors explain the fix. Distinctiveness: the concept explains the main color, type and motion choices · a signature element exists · nothing generic by accident. Web platform: MWG guides retrieved for every capability built · native elements before libraries · fallbacks match the browser-support policy · no global * resets or global reduced-motion kill switch. Codebase: no second design system; new values added as tokens in the existing convention; diff contains only in-scope changes.


8. Score, attribute and close out

Follow references/workflow/scoring-and-closeout.md. At Standard effort and above, write SCORECARD.md into the project (from templates/SCORECARD.md), not just a summary in chat:

  1. Scorecard built from the request: requirements coverage first, then the criteria that matter for this deliverable (design, usability, accessibility, function, responsiveness, distinctiveness, motion, sound, video). 0–10 each with evidence. Unverified lines are capped at 6.
  2. Target: overall ≥ 8 and no line < 6. Below target → list the 3 highest-impact fixes and offer /dent:improve.
  3. Attribution: summarise which key decisions came from DENT, from the user, from your own judgment, and any Overrides.
  4. Close-out: what was built, requirements met, checks run and results, deviations, known issues, next step (one line). If MWG was consulted, end the delivering response with the ### Modern Web Guidance disclosure (references/modern-web-guidance.md §2 step 5).
  5. Approval: end every delivered design with one question: "Approve it so I write DESIGN.md?" (or write DESIGN.md directly at All-out, marked Draft until approved). Then stop: no endless polishing.

9. Styles and reference map

9.1 Applying an aesthetic style

  1. Find the style in references/styles/_styles-index.md (name, alias or feeling words). Open only that style's file.
  2. Its Overrides replace the matching [Convention] defaults; anything it doesn't mention keeps the default.
  3. [Rule]s and gates are never overridden; each style's "Accessibility watch-outs" gives the fix.
  4. [Law]s still apply: a style changes how hierarchy looks, not whether there is hierarchy.
  5. The style's Signature moves must be present.
  6. Platform navigation and gestures stay native.
  7. One base style + at most one accent style. Never three. Record both in the Decision Log.

9.2 Reference map (open only what the task needs)

NeedOpen
Routing any request to the right commands and pipelinereferences/workflow/orchestration.md
Any HTML/CSS/client-side JS: Modern Web Guidance (primary web authority), precedence, DENT → MWG map, native-first, offline lookupreferences/modern-web-guidance.md (MWG skill itself: ../modern-web-guidance/SKILL.md)
Questions, effort levels, add-on routing, "don't ask" defaultsreferences/workflow-questions-and-effort.md
Task depth, PLAN.md, plan audit, approval, traceability, decisionsreferences/workflow/planning-and-plan-audit.md
Phased execution, scope protection, design-system inventory, definition of donereferences/workflow/execution-and-scope.md
Verification: functional, visual, responsive, a11y, performance, regression, journeysreferences/workflow/verification.md
Self-score, improve loop, attribution, close-outreferences/workflow/scoring-and-closeout.md
All-out effort: what it promises + exhaustive sweeps per deliverablereferences/workflow/all-out.md
Existing products: redesign, audit, adopt a codebasereferences/modes-existing-products.md
Step-by-step landing page / marketing sitereferences/recipes/landing-page.md
Step-by-step app screens, dashboards, flowsreferences/recipes/app-and-dashboard.md
Tokens and core component specsreferences/foundations-and-components.md
More components (date picker, upload, command palette, tables…)references/components-extended.md
Motion menu: which kinds of motion to ask about, intensities, minimum motion per effort levelreferences/motion/motion-menu.md
Interface motion, hover, scroll, depth, typography craftreferences/modern-layer.md
Concept, category cliché, composition, directions processreferences/creative-direction.md
Named creative directions (web, motion, sound, video)references/creative/directions-library.md
Hero titles and expressive typographyreferences/creative/hero-typography.md
Generic/AI-slop detection with reasonsreferences/creative/anti-slop.md
Aesthetic styles (41)references/styles/_styles-index.md + one style file
Brief, JTBD, metrics, experimentsreferences/design-brief-and-metrics.md
Choosing UI/component librariesreferences/catalog/component-selection.md
Library catalogue: primitives and full systemsreferences/catalog/components-primitives-and-systems.md
Library catalogue: collections and specialized componentsreferences/catalog/components-collections-and-specialized.md
Fonts, icons, images, 3D, textures, footage, sound and music sources + licencesreferences/catalog/asset-sources.md
Motion graphics, kinetic type, 1D/2D/2.5D/3D, storyboardsreferences/motion/motion-graphics.md
Animation and graphics technology catalogue + selectionreferences/motion/motion-tech-catalog.md
3D and real-time graphicsreferences/motion/3d-and-realtime.md
Motion language per project + motion verificationreferences/motion/motion-system-and-verification.md
Sound design: UI sound, sonic identity, ambience, music, interactivereferences/audio/sound-design.md
Narration, voice-over, TTS, captionsreferences/audio/voice-and-narration.md
Audio technology and asset cataloguereferences/audio/audio-tech-catalog.md
Audio implementation and verificationreferences/audio/audio-implementation-and-verification.md
Video types, scripts, storyboards, craftreferences/video/video-types-and-craft.md
Video tools (generative + code-based + editors), pipeline, export specsreferences/video/video-tools-and-pipeline.md
Landing, pricing, dashboard, onboarding, auth, forms, checkout, settings, AI/chat UIreferences/page-and-flow-patterns.md
Teams, billing, roles, notifications, docs, blog, 404, legalreferences/complex-product-patterns.md
iOS, Android, web, desktop, watch, TV, car, voice, XR, hapticsreferences/platforms-and-devices.md
Industry needs (fintech, health, commerce…)references/industry-playbooks.md
Color, typography, data viz, icons/imagery, brand, i18n, UX writingreferences/color.md, references/typography.md, references/data-visualization.md, references/icons-illustration-and-imagery.md, references/brand-identity.md, references/internationalization.md, references/ux-writing.md
Favicons, app icons, OG images, social, print, slides, emailreferences/assets-and-marketing-graphics.md
Research, personas, IAreferences/research-and-information-architecture.md
Persuasion and deceptive-pattern checksreferences/persuasion-and-ethics.md
Accessibility in depthreferences/accessibility.md
Tokens architecture and code per stackreferences/design-tokens-and-code.md
Visual review loop, rubric, optical polishreferences/quality-review.md
Previews, Figma, Storybook, handoffreferences/tooling-and-handoff.md
Reference sites and inspirationreferences/inspiration-and-reference-sites.md
DESIGN.mdreferences/design-md-template.md
Templatestemplates/PLAN.md, templates/AUDIT.md, templates/REVIEW.md, templates/SCORECARD.md, templates/MOTION.md, templates/SOUND.md, templates/VIDEO.md
Checks you can runscripts/verify_page.py, scripts/contrast.py, offline MWG lookup scripts/mwg.py; plugin updater scripts/update_plugin.py

To extend DENT: new styles follow references/styles/_style-template.md + a row in the index; new catalogue entries follow references/catalog/README.md; new topics get a reference file, a row above, and a row in the add-on routing table.

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.