compound-knowledge-work/tests/fixtures at main · RDEL-Group/compound-knowledge-work · GitHub
Skip to content

Latest commit

 

History

History

fixtures

README.md

Import-pipeline test fixtures

Small reference documents used to verify the import pipeline (skills/import-summarizer/scripts/convert_document.sh + the PDF Read-tool path). Consumed by the A7 CI smoke test.

Every fixture contains the sentinel string CKW_FIXTURE_SENTINEL and the line "The FY2026 pacing target is 92 percent." so a converter's output can be checked for real extracted text (not silent garbage).

Fixture Purpose Documented conversion path
fixture.docx Word extraction convert_document.sh → pandoc/textutil/python-docx
O'Brien Brief.docx Apostrophe in filename (was a shell-injection / syntax-error bug) same
Report.DOCX Uppercase extension (previously errored) same (extension matched case-insensitively)
fixture.pptx PowerPoint (previously produced empty output via textutil) convert_document.sh → pandoc ≥ 3.7
fixture.html HTML convert_document.sh → pandoc/textutil
fixture.rtf RTF (supported by script, previously undocumented) convert_document.sh → textutil/pandoc
fixture.txt Plain text direct read
fixture.pdf PDF (previously dumped raw bytes via textutil) Claude's native Read tool — NOT the script; the script refuses .pdf and points at the Read tool

Regenerating

# docx / pptx via pandoc; pdf via macOS cupsfilter; rtf via textutil
pandoc src.md    -o fixture.docx
pandoc slides.md -o fixture.pptx
cupsfilter fixture.txt > fixture.pdf
textutil -convert rtf -output fixture.rtf fixture.txt

Acceptance check

Each non-PDF fixture must yield non-empty output containing the sentinel when passed to convert_document.sh; .pdf must make the script exit non-zero with a message pointing at Claude's Read tool. See A7 (.github/workflows/smoke.yml).