A Claude Code skill that acts as a quality gate for rendered document deliverables — PDF, DOCX, memos, proposals, reports, SOPs — and for rendered charts, org charts, and diagrams.
Plenty of skills fight the generic-AI look in web UI (purple gradients, three-card layouts, eyebrow labels). Documents have their own AI tells, and almost nothing polices them. The strongest one: the tinted callout box with a colored accent bar and a bold lead-in label. Readers now recognize the box itself — before reading a word — as machine styling.
This skill is the paper-side counterpart of the web design-taste skills: a short list of banned layout devices, the native print devices to use instead, structural rules for files recipients will edit, and an exit checklist that includes rasterizing and eyeballing every page before anything ships.
mkdir -p ~/.claude/skills/document-design
curl -o ~/.claude/skills/document-design/SKILL.md \
https://github.com/ghraw/WolfurX/document-design/main/SKILL.mdOr clone and copy. Claude Code picks it up automatically; it triggers whenever a document template, renderer, or rendered deliverable is being written, edited, or sent.
A rendered chart is a document, not an app screen. The skill separates encoding (a cluster tint, a dashed outline for a seat that does not exist yet) from decoration (an accent bar on every card in a tier, drop shadows, counts rendered as filled pills) and cuts only the second. Watch inherited renderers in particular: a chart drawn before you adopted a layout gate carries its tells forward every time you evolve it.
Accent-bar callout boxes, emoji headings, bold-label lists, cards and shadows on paper, full-grid heavy tables, status badges, centered body text, and over-sectioned heading-per-paragraph outlines — replaced by prose, indents, footnotes, thin rules, and hierarchy built from size, weight, and space.
MIT