- claude contributed src
  - processes zip from url using (system
    installed) curl for download

- claude contributed src
  - Opens the zip with std.zip.ZipArchive (reads the whole file into
    memory)
  - Locates pod.manifest inside the archive to discover document paths
    and languages
  - Extracts markup files (.sst/.ssm/.ssi) as in-memory strings
  - Extracts images as in-memory byte arrays
  - Extracts conf/dr_document_make if present
  - Presents these to the existing pipeline as if they were read from
    the filesystem
  - Some security mitigations:
    - Zip Slip / Path Traversal: Reject entries containing `..` or
      starting with `/`; canonicalize resolved paths and verify they
      fall within extraction root
    - Zip Bomb: Check `ArchiveMember.size` before extracting; enforce
      per-file (50MB) and total size limits (500MB)
    - Entry Count: Limit number of entries (a pod should have at most
      ~100 files)
    - Path depth: limit (Maximum 10 path components).
    - Symlinks: Verify no symlinks in extracted content before
      processing (post-extraction recursive scan)
    - Filename Validation: Only allow expected characters; reject null
      bytes
    - Malformed Zips: Catch `ZipException` from `std.zip.ZipArchive`
      constructor
    - Cleanup on error

- FIXES issue with .tex files and xetex finding image paths when run
  within latex/ output directory

- revisit links (fix later)

- preferable, endnote parent object number
  available for use (as here in text output,
  compare "endnotes, add caller ocn" commit)

- spine --text [--output=output path] [markup source]