agentleFS
S3n0001S3n0001/agent-skills/craft-contextual-ui/SKILL.md

craft-contextual-ui

Skill1 starsChanged 2 months agoMITView on GitHub

Design, critique, or refine product-specific frontend UI for hierarchy, responsive and accessible states, and rendered polish, verified in a browser. Code runs under engineer-code-change.

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 9 are distinctive, and 0 are 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.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read repository, product, brand, and accessibility guidance.

    line 18
  2. Distinctive:

    Every visible choice should clarify importance, action, grouping, state, or identity.

    line 26
  3. Distinctive:

    Use real content. Write specific labels, stable action names, useful empty states, and errors that explain recovery.

    line 29
  4. Distinctive:

    Use semantic elements, associated labels, logical headings, visible focus, keyboard-operable controls, sufficient contrast, and …

    line 32
  5. Distinctive:

    Use motion to explain change or reinforce one moment, not as scattered decoration.

    line 33
  6. Distinctive:

    Check consumers before changing shared or global styles.

    line 34
  7. Distinctive:

    Check hierarchy, alignment, rhythm, density, contrast, clipping, wrapping, overflow, real content, and fit with adjacent surfaces.

    line 44
  8. Distinctive:

    When implementation is authorized, fix visible and interaction defects through the code-change workflow, rerun affected non-visual checks, …

    line 46
  9. Distinctive:

    no runnable browser or representative state exists and the request needs visual verification.

    line 53

≈ “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

Craft Contextual UI

Own the interface decisions that make a surface fit its product, and the rendered evidence that they work. When files must change, engineer-code-change owns integration and completion unless the user invoked this skill as primary; either way this skill supplies the UI acceptance criteria and the browser loop.

Choose the mode first:

  • Critique or audit: inspect the rendered interface and return prioritized, evidence-backed findings without editing.
  • Design only: produce the direction or artifact without implying it is implemented.
  • Authorized implementation: make the scoped UI changes through the code-change workflow, then validate the rendered result here.

Ground the interface

  1. Read repository, product, brand, and accessibility guidance.
  2. Inspect the current rendered surface when possible, plus nearby screens, components, tokens, assets, content patterns, and breakpoints.
  3. Identify the audience, the surface's primary job, the dominant action, what must be noticed first, and what should stay familiar.
  4. For an established product, extend its visual language. For greenfield or aesthetic-direction work, infer a concrete product world from the brief and use the aesthetic direction reference to avoid default looks.
  5. Form one design thesis: hierarchy, layout logic, type and color roles, interaction character, and at most one or two distinctive moves rooted in the subject. Commit to it unless alternatives were requested.

Shape the experience

Every visible choice should clarify importance, action, grouping, state, or identity.

  • Reuse existing primitives, tokens, icons, and dependencies; add a primitive only when the system cannot express the pattern cleanly.
  • Use real content. Write specific labels, stable action names, useful empty states, and errors that explain recovery.
  • Design the applicable loading, empty, error, disabled, validation, hover, focus, active, selected, overflow, and long-content states.
  • Recompose at narrow widths instead of shrinking. Preserve reading order, content priority, touch targets, and freedom from horizontal overflow.
  • Use semantic elements, associated labels, logical headings, visible focus, keyboard-operable controls, sufficient contrast, and reduced-motion behavior.
  • Use motion to explain change or reinforce one moment, not as scattered decoration.
  • Check consumers before changing shared or global styles; keep DOM and styling complexity proportional to the idea.

Reject interchangeable output: repeated rounded containers without grouping value, fake metrics, decorative charts, arbitrary gradients or glass effects, generic marketing copy, and a familiar type-and-color recipe applied regardless of product. Use such a pattern only when the content or the established system makes it the clearest choice.

Prove it in the browser

Source inspection is not visual verification.

  1. Launch the real interface through the repository's supported path and exercise the primary flow.
  2. Inspect representative desktop and mobile widths and any intermediate width where layout risk concentrates.
  3. Check hierarchy, alignment, rhythm, density, contrast, clipping, wrapping, overflow, real content, and fit with adjacent surfaces.
  4. Exercise the relevant states and failure behavior. Navigate by keyboard; inspect focus, labels, and reduced motion.
  5. When implementation is authorized, fix visible and interaction defects through the code-change workflow, rerun affected non-visual checks, and repeat the rendered pass. Otherwise report the defects.

For a subjective, ambitious, or high-risk surface, use a fresh evaluator given the product job and an observable rubric, not the builder's rationale, and validate its evidence before revising.

Stop and ask when

  • a product-defining choice (brand direction, information architecture, primary action) cannot be inferred safely;
  • no runnable browser or representative state exists and the request needs visual verification;
  • the fix requires changing shared styles whose consumers are outside the requested scope.

Done when

The primary flow and relevant states were exercised in a real browser at representative widths with keyboard access checked, or the missing proof is named explicitly (never call it visually verified otherwise). Report the design direction, surfaces changed or reviewed, widths and states exercised, defects corrected, unverified states, and residual accessibility or product-fit risk.

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.