agentleFS
onepurponepurp/Dembrandt-Skill/SKILL.md

dembrandt

Skill1 starsChanged 3 months agoMITView on GitHub

Extract and analyze live website design systems into design tokens, brand identity, typography, colors, spacing, surfaces, components, WCAG reports, DESIGN.md, Tailwind v4 themes, DTCG JSON, drift reports, and HTML or brand-guide artifacts. Use when asked to reverse-engineer, document, compare, audit, or enforce a website’s visual system from a URL, or when Dembrandt CLI/MCP workflows are needed.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. dembrandt/dembrandt-skills★ 194same 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.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 4 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use Dembrandt to measure a website’s observed design system rather than inventing one.

    line 8
  2. Very specific:

    Prefer the Dembrandt MCP tools: when they are available in the current agent environment.

    line 14
  3. Distinctive:

    Use the CLI through the shell: when MCP is unavailable, when files must be written locally, or when a reproducible CI command is required.

    line 15
  4. Distinctive:

    Read references/mcp-tools.md for the MCP tool contract, references/cli-and-exports.md for CLI options and output selection, and …

    line 17
  5. Very specific:

    Select the observation context: Use desktop by default. Add mobile when responsive behavior matters, dark-mode when the dark theme is in …

    line 22
  6. Very specific:

    Extract machine-readable evidence: Prefer --json-only for agent processing and --save-output when a durable artifact is needed.

    line 24
  7. Distinctive:

    Verify the result: Check that the output exists, is parseable, and matches the requested scope.

    line 27

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Dembrandt

Use Dembrandt to measure a website’s observed design system rather than inventing one. It renders the target with a real browser, reads computed DOM styles, and returns evidence for colors, typography, spacing, borders, radii, shadows, motion, components, responsive behavior, branding, frameworks, and icons.

Authorization rule: Analyze only sites the user owns or is authorized to inspect, and only where automated access is permitted. Respect the site’s Terms of Service, robots.txt, rate limits, copyright, and third-party brand rights. Treat cookies, API keys, and extracted brand assets as sensitive.

Choose the execution route

  1. Prefer the Dembrandt MCP tools when they are available in the current agent environment. Use the extraction tools for live pages, then poll the returned job with get_job_status unless the user explicitly needs a blocking result.
  2. Use the CLI through the shell when MCP is unavailable, when files must be written locally, or when a reproducible CI command is required. Use npx dembrandt ... for an ephemeral install or the globally installed dembrandt command for a project that already depends on it.
  3. Use pure analysis tools such as compute_drift, get_findings, export_dtcg, or generate_design_md on an existing extraction instead of reloading the website.
  4. Read references/mcp-tools.md for the MCP tool contract, references/cli-and-exports.md for CLI options and output selection, and references/ci-drift.md for baseline and CI work.

Standard extraction workflow

  1. Define the question and scope. Decide whether the user needs a full design system, a focused category, a responsive comparison, an accessibility audit, a brand guide, or a drift check. Confirm the URL and authorization when it is not obvious.
  2. Select the observation context. Use desktop by default. Add mobile when responsive behavior matters, dark-mode when the dark theme is in scope, wcag for contrast evidence, and crawl or explicit paths when one page is insufficient. Use sitemap when navigation links are incomplete.
  3. Prepare the browser. Dembrandt requires Node.js 18 or later and a browser binary matching its bundled playwright-core. Run dembrandt install-browser once before the first CLI extraction. In Linux CI, install the matching browser system dependencies or use a compatible Playwright container. If a browser already exists, use BROWSER_CDP_ENDPOINT with Chromium.
  4. Extract machine-readable evidence. Prefer --json-only for agent processing and --save-output when a durable artifact is needed. Use --slow for heavy single-page applications. Preserve the raw extraction before transforming it.
  5. Interpret conservatively. Distinguish observed values from recommendations. Treat confidence, frequency, semantic context, viewport, color scheme, and page coverage as evidence. Do not fill omitted sections with guessed defaults. Account for the limitations below before making design claims.
  6. Export for the destination. Choose DTCG for token pipelines, DESIGN.md for AI-readable project guidance, Tailwind v4 @theme CSS for a Tailwind project, --brand-guide for a printable summary, and --html or render_report for a self-contained review artifact.
  7. Verify the result. Check that the output exists, is parseable, and matches the requested scope. For a comparison, use the same browser, viewport, routes, wait behavior, and Dembrandt version for baseline and candidate whenever possible.

Quick command selection

NeedPreferred command or operationResult
Full extractiondembrandt URL --json-onlyRaw extraction JSON on stdout
Persist JSONdembrandt URL --save-outputTimestamped JSON under output/
W3C token pipelinedembrandt URL --dtcg.tokens.json DTCG export
AI-readable design guidancedembrandt URL --design-mdDESIGN.md
Tailwind v4 projectdembrandt URL --tailwindObserved-value @theme CSS
Accessibility reviewdembrandt URL --wcagDOM-based contrast results and grades
Responsive or theme comparisonAdd --mobile or --dark-modeContext-specific extraction
Multi-page systemdembrandt URL --crawl N or explicit pathsMerged extraction with cross-page evidence
Printable summarydembrandt URL --brand-guideBrand-guide PDF
Drift gatedembrandt URL --compare baseline.json --json-onlyDrift JSON and exit status

Decision rules and failure handling

Use --slow before concluding that a JavaScript-heavy site is inaccessible. If Chromium is blocked or unreliable, install Firefox and retry with --browser=firefox; use --stealth only when the user is authorized to access the site and the site’s rules permit it. Use --no-sandbox only in an appropriately isolated container or CI runner. For authenticated pages, pass a cookie only when the user has supplied an authorized session value; never expose it in logs or final output.

Treat extraction failure as different from design drift. In CLI drift gates, exit code 1 means drift, 2 means extraction or browser failure, and 67 means a navigation timeout that is generally retryable with --slow. Do not approve or overwrite a baseline merely because extraction failed or because the candidate used a different observation context.

Dembrandt does not fully analyze Canvas/WebGL-rendered interfaces, may miss dynamically loaded content, does not automatically infer dark mode, and reads hover/focus states primarily from CSS rather than complete interactive behavior. The default desktop viewport is 1920×1080; the mobile profile is 390×844. Report these constraints whenever they could affect the user’s conclusion.

MCP operating pattern

Extraction MCP tools return a job_id by default because browser work can take 15–40 seconds. Poll get_job_status until the status is completed, then pass the returned extraction to the pure analysis or export tools. Set sync: true only when blocking is acceptable and the result is small enough to use directly. Keep concurrent live extractions modest; the server limits active jobs to two.

Use focused tools when the user asks for one category, and use get_design_tokens when the complete context is needed. After extraction, prefer this sequence for quality work: get_findings to identify problems, compute_drift to compare against a baseline, then export_dtcg, generate_design_md, or render_report for delivery. See references/mcp-tools.md for the exact catalog.

Output discipline

Label every deliverable with its source URL, extraction context, timestamp, and whether it is raw evidence or an interpretation. Keep raw JSON alongside derived files. When summarizing colors or typography, retain the original notation and semantic context even if the user requests a transformed format. Tailwind output contains observed values only: do not present absent shade ramps, hover variants, or on-color values as if Dembrandt measured them.

References

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.