d2-slides

Write a Markdown file with slides separated by lines of exactly --- and each figure as a ```d2 fence, then run build-deck.mjs with the file and an output directory. It compiles each figure with D2's TALA layout and emits index.html, print.html, per-slide SVGs, a copy manifest, and slides.json.

Run vlmkit check integrity on the built index.html across viewports and on print.html at 1280px, which reads every slide at once. Then run check copy with the generated copy.txt manifest to catch lines the fixed 1280x720 frame cut off.

The d2 CLI must be installed and d2 layout must list tala as bundled. The builder itself has no npm dependencies, and the generated deck loads no network resources, so D2 with TALA is the only hard requirement.

This is a known vlmkit gate limitation: a manifest line whose rendered text is assembled from several inline children, such as code or strong tags, can read as invisible. Re-run with --allow-invisible unknown and check the missing count, which is the part that matters.

Do not use it for a single standalone diagram, which belongs in the d2-diagram workflow, or for animated figures, which belong in explanatory-animation. It targets whole decks presented on screen or printed to PDF.

Run d2-facts.mjs with --from-svg on the built slide SVG and --expect pointing to a fact sheet listing required boxes, edges, and forbidden connections. It exits 1 and names the box or edge that is wrong, catching errors the page gates cannot see.