v0.14.0

Minor Changes

• 604ca24: serve: add --host / SIDESHOW_HOST to bind one address

  serve listened on every interface with no way to restrict it, and printed
  listening on http://localhost:PORT regardless — so a server sharing a host or
  a network with anything else was reachable from it, and the startup line said
  otherwise. The default is unchanged (every interface, which is what containers
  and LAN-shared instances need); --host 127.0.0.1 now keeps it off the network
  entirely, and the startup line reports the address actually bound.

• 14a8ea3: viewer: fold the card's export actions into one share menu, with copy-as-markdown

  A card's footer carried three separate icons that all meant "take this
  elsewhere" — copy link, open in a new tab, open as a PNG — and no room for a
  fourth. They are now rows in a single Share menu, joined by Copy as
  markdown: the whole post as portable markdown, with prose kept as prose,
  code/diffs/terminal output/JSON/mermaid as fenced blocks, images as image links,
  and an html surface degraded to a link back to it rather than a dump of its
  markup.

  The flattening is served, not derived in the browser, so every tier can have it:
  GET /api/posts/:id/markdown returns the same text for curl and the CLI.

Patch Changes

• 68531b2: Show a recent-posts Home view for first-time visitors to multi-session workspaces, with live updates and safe themed previews.
• 11c4434: Cache fully pinned social-preview screenshots at the Cloudflare edge so repeated link unfurls avoid redundant Browser Rendering calls. Access, post existence, and revision are revalidated on every edge request, and token-protected boards keep private client cache headers.
• feac177: Make recent-post ordering deterministic when multiple writes share the same millisecond timestamp, preventing different SQLite versions from selecting different posts at the result limit.
• 102990f: Load the rich-surface renderers and publish-time parsers on first use instead of
  at boot. shiki, @pierre/diffs, markdown-it and @mermaid-js/parser are now
  imported when a markdown/code/diff/terminal surface is rendered or a diff/mermaid
  surface is published, rather than by every server at startup. Idle memory drops
  from ~132 MB to ~102 MB and boot time from ~486 ms to ~275 ms; a workspace that
  only ever uses html surfaces never loads them at all. No behavior change.
• ea2fb02: Highlight code against one theme instead of two when the color scheme is already
  known. shiki tokenizes once per theme, so asking for a light/dark pair costs
  exactly twice as much — and the viewer always tells the server which scheme it
  resolved, so half that work was being discarded with CSS. Rendering a
  1400-line code surface drops from 1.14s to 543ms and its document from 751KB to
  529KB; markdown with fenced code improves by roughly the same proportion. An
  unpinned direct load of /s/:id still gets both themes and follows the OS.
• b0f6a94: Comment authors are now derived from the session agent for CLI, MCP, and other programmatic writes. The reserved user label is limited to same-origin viewer comments, preventing agent integrations from forging user feedback.
• 6c1073a: Invalid Mermaid submitted through the posts API now returns a typed validation
  error with the failing request field, diagram type, complete parser diagnostic,
  and concrete retry steps. The existing human-readable error string remains
  for compatibility, and rejected diagrams are not persisted.