| name | verify-hebrew-render |
| description | Check that a Hebrew PDF actually rendered correctly — right font embedded, no silent substitution, direction sane — before it is sent to anyone. Use after generating any Hebrew PDF and before any send. Triggers on "check this Hebrew PDF", "did the font work", "is this letter rendered right". |
Verify a Hebrew render before it goes out
Every failure mode in this toolchain is silent. Run this rather than trusting
that the command succeeded.
python3 ${CLAUDE_PLUGIN_ROOT}/scripts/verify_render.py letter.pdf --expect-font DavidLibre
Add --json for machine-readable output. Exit status is 0 only when no problems
were found.
What it catches, and why each matters
| Signal that lies | What actually happens |
|---|
| Typst exit code | Emits warning: unknown font family and compiles anyway, substituting a Latin serif. Exit 0, PDF produced, Hebrew in the wrong face. |
| WeasyPrint exit code | Ignores unicode-bidi: plaintext with a warning; LTR paragraphs render inside an RTL document. |
pdftotext output | Reports logical order with embedding marks. A correct RTL render looks broken; a substituted-font render extracts perfectly. |
fc-match | Answers with its best substitute rather than failing when a font is absent. |
A substituted-font PDF still extracts every Hebrew codepoint correctly. Text
extraction cannot detect it. Only the embedded font list can.
What it cannot check
Text direction. There is no reliable way to read paragraph direction back out
of a PDF. The script says so rather than pretending otherwise. Finish the job by
looking:
pdftoppm -png -r 100 -f 1 -l 1 letter.pdf /tmp/page
Confirm Hebrew is right-aligned, any Latin paragraph is left-aligned, and
Name <address> pairs and numbers are intact. Report what you saw, not that the
command exited cleanly.
