/cs-md-document

disable-model-invocation: true
description: Convert long-form markdown (specs, RFCs, reports, plans, explainers) into a single-file interactive HTML document. Runs the md-document pipeline (parser → renderer → injector) with the user's design-system tokens. Sticky TOC, search filter, code-copy buttons, scrollspy, WCAG-AA palette. Refuses input < 100 lines (Shihipar) or unfinished onboarding. Output is a single .html file (Google Fonts + Prism.js CDN as the only externals).
argument-hint: "<path to markdown file> [--out <dir>] [--features search,copycode,smoothscroll,scrollspy]"

/cs-md-document — Long-form markdown → HTML

Convert the markdown at **$ARGUMENTS** into a single-file interactive HTML document.

Pre-flight gates (refuse, never override)

1. **Input < 100 lines** → refuse (markdown wins below the threshold per Shihipar). `wc -l <path>` to confirm. 2. **Design-system not onboarded** → refuse, surface `/cs-design-system`. 3. **Output directory unwritable** → refuse, ask user for an alternate via `--out`.

Pipeline

# 1. Classify (if not already routed by orchestrator)
python3 ${CLAUDE_PLUGIN_ROOT}/skills/markdown-html-orchestrator/scripts/doctype_classifier.py \
    --input "<path>.md" --output json \
  | python3 ${CLAUDE_PLUGIN_ROOT}/skills/markdown-html-orchestrator/scripts/route_explainer.py

# 2. Resolve the output path
python3 ${CLAUDE_PLUGIN_ROOT}/skills/markdown-html-orchestrator/scripts/output_path_resolver.py \
    --input "<path>.md" --doctype document

# 3. Parse → render → inject
python3 ${CLAUDE_PLUGIN_ROOT}/skills/md-document/scripts/markdown_parser.py \
    --input "<path>.md" --output /tmp/sections.json
python3 ${CLAUDE_PLUGIN_ROOT}/skills/md-document/scripts/html_renderer.py \
    --sections /tmp/sections.json --output <resolved-out>.html
python3 ${CLAUDE_PLUGIN_ROOT}/skills/md-document/scripts/interactivity_injector.py \
    --file <resolved-out>.html \
    --features search,copycode,smoothscroll,scrollspy

What ships in the HTML

• Sticky-sidebar TOC (default; configurable via `toc.behavior` in design-system)
• Scrollspy: `aria-current="location"` on the TOC entry for the section in view
• Search bar (Esc clears): filters which H2 sections are visible
• Code-copy buttons on every `<pre>` (vanilla `navigator.clipboard` with `execCommand` fallback)
• Smooth-scroll on TOC link clicks
• Prism.js syntax highlighting (autoloader fetches only the languages this doc uses)
• 12 brand CSS custom properties from the design-system's `derived_palette`
• `@media (prefers-reduced-motion: reduce)` honored
• Print-friendly via the browser's native print stylesheet (no `@page` overrides needed for documents)

Hard rules

• Output is one `.html` file. No multi-file output, no extracted CSS/JS, no asset folders.
• External CDN: `fonts.googleapis.com` + `cdn.jsdelivr.net` (Prism). Nothing else.
• No JS framework runtime. Vanilla JS + IntersectionObserver only.
• Re-running on the same input writes `doc-{slug}-2.html` etc. (collision suffix).

Output

Returns: input lines, output path, design style applied, top 3 features used, one forcing question.

References

See `markdown-html/skills/md-document/references/`:

• `information_density_patterns.md` — Shihipar + Tufte + Wattenberger
• `toc_and_nav_ux.md` — NN/g + WCAG + ARIA
• `single_file_html_discipline.md` — Single-file artifact rationale