fix: keep generated Markdown content-only

[parvahuja]

What changed

• Remove page-level MDX imports, exports, and presentation-only elements from generated Markdown.
• Keep the same text unchanged when it appears inside a fenced code example.
• Extend the post-build audit so CI rejects these nodes or any remaining executable MDX expressions if they reach a .md page or llms-full.txt.
• Update the Vocs preview to wevm/vocs#598, which preserves valid content after nested expressions.

Why

The generated .md pages and llms-full.txt are content interfaces. They should contain documentation and code examples, not executable page source or browser-only metadata.

The Vocs update fixes a separate tree-editing bug: removing an expression nested inside an element could delete the document's final content block. Tempo's AI guide ended with a resource table, so that table could disappear from generated Markdown.

The Tempo transform runs only on generated Markdown. It does not change the source MDX or visible docs content.

Validation

• env CI=true pnpm test — 300 tests passed
• env CI=true VITE_USE_HTTP=true NODE_OPTIONS=--max-old-space-size=4096 pnpm build
• pnpm run check:markdown
• ./node_modules/.bin/biome check src/lib/markdown-output.ts src/lib/markdown-output.test.ts scripts/check-markdown-components.mjs .github/workflows/verify.yml
• Verified the generated AI guide retains both llms.txt links and no longer contains its style block
• Verified generated token-list and private-zone pages no longer contain page-level imports or exports

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
tempo-docs	Ready	Preview, Comment	Jul 25, 2026 12:56am