agentleFS
swyxioswyxio/skills/design-apps-with-imagegen/SKILL.md

design-apps-with-imagegen

Skill176 starsChanged 7 months agoMITView on GitHub

Explore and implement app or site redesigns through free Mobbin reference research, four distinct ImageGen directions, user selection or delegated choice, and matched visual critique. Use for image-led design exploration, not routine UI fixes or implementation of an already selected design.

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 6 are distinctive, and 0 are 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.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read imagegen-workflow. Use Chrome computer use to search free Mobbin before ideation, then generate four distinct ImageGen directions …

    line 18
  2. Distinctive:

    Use representative content and controls.

    line 20
  3. Distinctive:

    Ask once when the user has not selected or delegated the choice.

    line 26
  4. Distinctive:

    Use an isolated prototype for structural or behavior-heavy exploration.

    line 32
  5. Distinctive:

    Use visual-comparison to inspect matched reference and implementation screenshots.

    line 42
  6. Distinctive:

    Stop when the requested flow works, scoped contracts remain intact, and the chosen design's defining traits survive at relevant sizes.

    line 48

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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 Apps with ImageGen

Explore composition and interaction through images, implement the chosen direction, and compare the actual product against it.

1. Establish scope

Inspect the affected screens, real content, capabilities, state owners, and responsive behavior. Separate existing contracts from proposed changes. Preserve scoped functionality and data unless the user authorizes changes; generated controls and text are proposals, not product truth.

Retain applicable navigation, account controls, and workspace access through loading and recovery states. Follow design-preferences; shell replacement must be in scope. Report unrelated layout gaps rather than expanding the redesign.

2. Research and explore

Read imagegen-workflow. Use Chrome computer use to search free Mobbin before ideation, then generate four distinct ImageGen directions including a purposeful wildcard. Give each its own creative brief and reference set. Differences must include composition or interaction, not just palettes and corner radii.

Use representative content and controls. For responsive products, show desktop and compact compositions; include expanded menus or settings when they determine the information architecture. Check feasibility and capability coverage before presenting directions.

3. Resolve the direction

Present A–D with images, a design thesis, advantage, tradeoff, and recommendation. Bundle unresolved material behavior choices into the same selection question.

Ask once when the user has not selected or delegated the choice. Honor existing authorization, “pick the defaults,” and explicitly requested combinations; restate a hybrid without asking again unless a consequential ambiguity remains. Silence is not selection.

Record the chosen direction, authorized behavior changes, and three defining traits including its signature move. These are the acceptance reference.

4. Implement

Use an isolated prototype for structural or behavior-heavy exploration; bounded changes can be implemented directly. Keep prototype network effects, persistence, analytics, and destructive actions isolated from live data.

Translate the design into explicit geometry: dominant region, fold, column widths, copy measures, scroll/sticky regions, and breakpoint disclosure. Use representative long text and density early; do not let wrapping or familiar stacked cards replace the chosen composition.

Generate production assets when texture, illustration, or organic detail defines the direction. Use code for dynamic text, interactive states, and simple geometry. Prove an asset-dependent treatment at its actual display size before building the full set.

Consult relevant rows of the miss ledger when a known implementation failure applies; it is not a second checklist.

5. Compare, correct, and integrate

Use visual-comparison to inspect matched reference and implementation screenshots. Verify the defining traits, hierarchy, density, typography, and responsive composition. Builds and DOM assertions cannot establish visual fidelity.

Fix material differences and recapture affected states and sizes. Do not require a correction round when the evidence already supports completion. Return to the user only if resolution needs an unapproved change to the selected direction or behavior.

Exercise the requested flow and affected capabilities after integration. Use the edge-case matrix selectively for changed layout or behavior; broader redesigns warrant broader coverage. Check alternate themes when supported, reachable controls, overflow, and preservation of input, selection, focus, and scroll across layout changes. Serious failures that prevent the scoped flow are blockers.

Stop when the requested flow works, scoped contracts remain intact, and the chosen design's defining traits survive at relevant sizes. Document acceptable adaptations; do not keep gathering proof for unrelated improvements. Follow repository rules and existing user authorization for merge and deployment.

Handoff

Retain inspected reference links, the selected mock, capture conditions, final screenshots, and any material deviations in the established design-history location. Keep unused explorations outside the repository.

Report the chosen direction, reference principles, behavior changes, visual evidence, remaining limitations, asset paths, and source-control state. Propose a generalized ledger lesson only when a useful new failure mode emerged. Running this skill does not authorize editing the skill or its ledger.

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.

Reviews can't be read right now.

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.