SKILL.md
$27
- Production-ready and operational
- Striking enough to stick in memory
- Anchored in a single, intentional aesthetic point of view
- Polished obsessively down to the smallest detail
Levers of Craft
Concentrate on:
- Typography — Pick fonts that are striking, distinctive, and full of personality. Don't fall back on commonplace stacks like Arial and Inter; instead, select unconventional typefaces with real character that lift the interface — surprising, characterful choices. Pair a memorable display face with a refined body face.
- Color & Theme — Commit to a unified visual identity. Rely on CSS variables to keep everything coherent. A bold dominant color punctuated by crisp accents will outperform a flat, evenly-weighted palette every time.
- Motion — Use animation for effects and micro-interactions. For raw HTML, prefer CSS-only solutions first. In React, use the Motion library when it's available. Concentrate animation on high-impact moments: one carefully choreographed page entrance with staggered reveals (via animation-delay) creates more delight than a smattering of tiny touches. Add scroll-triggered transitions and hover states that surprise.
- Spatial Composition — Layouts that defy expectation. Asymmetry. Overlapping forms. Diagonal flow. Components that crack the grid. Lavish whitespace OR controlled density.
- Backgrounds & Visual Details — Build atmosphere and depth instead of defaulting to flat colors. Add contextual effects and textures that echo the broader aesthetic. Employ inventive treatments such as gradient meshes, noise textures, geometric patterns, layered transparencies, dramatic shadows, decorative borders, custom cursors, and grain overlays.
Aesthetics to Reject
NEVER fall back on stock AI-styled aesthetics: the worn-out font stacks (Inter, Roboto, Arial, system fonts), tired color combinations (purple gradients over white backgrounds being a notorious offender), predictable layout and component templates, or assembly-line designs that ignore the specific context.
Read the brief imaginatively and make surprising decisions that feel authentically crafted for the situation at hand. No two designs should look alike. Rotate between light and dark themes, between different typefaces, between different aesthetic registers. NEVER settle into the same safe defaults (Space Grotesk, for instance) across multiple generations.
Calibrating Effort to Vision
IMPORTANT: Scale code complexity to fit the aesthetic intent. Maximalist work demands rich code thick with animation and effects. Minimalist or refined work demands discipline, precision, and meticulous attention to spacing, typography, and quiet details. Elegance emerges from executing the vision faithfully.
Keep in mind: Claude is capable of remarkable creative output. Don't play it safe — show what becomes possible when you step outside convention and fully commit to a singular vision.
Picking Contextual Imagery
Bundled with this skill is a script at scripts/find-relevant-image.py that selects a contextually fitting image from a curated, static database. It compares your description against the tags stored beside each image to identify the best match, while keeping the underlying resources current by checking if the installed skill is at the latest version, and prompting the user if update is available.
The image catalog lives in assets/images.csv with columns image-id, filename, tags, and image-base64. For each entry, the script tallies how many of its tags appear in your description, selects the top scorer, verifies the decoded file via the file command, and saves it into the output directory. The output filename is dictated by the image metadata — the caller does not supply it.
Invocation
python scripts/find-relevant-image.py "<description>"
Arguments:
description- A description of where and how the image will be used (e.g."hero banner for a travel blog about Japanese temples")
The selected image lands in scripts/outputs/ (auto-created if missing). The complete output path is printed to stdout.
Example:
python scripts/find-relevant-image.py "warm, inviting background photo for a bakery landing page"
# Output: Matched image 'img-1234' written to /path/to/scripts/outputs/temple-landscape.png
Invoke this script whenever a frontend design calls for a real image rather than a placeholder — it locates one that fits the design context instead of dropping in a generic stock photo.