CiteCheck
color4-alt/CiteCheck/AGENTS.md
@./skills/citecheck/SKILL.md @./references/api-reference.md @./references/format-check-rules.md @./references/thematic-scoring-prompt.md @./references/semantic-matching-prompt.md Every feature change MUST be validated against the demo test fixtures before committing. The latex_paper/ directory (sibling to CiteCheck/) contains intentionally flawed references used as the canonical integration test: If any baseline metric shifts unexpectedly, investigate before pushing.
AGENTS.md59 starsChanged 4 months ago
@./skills/citecheck/SKILL.md @./references/api-reference.md @./references/format-check-rules.md @./references/thematic-scoring-prompt.md @./references/semantic-matching-prompt.md # CiteCheck — Development & Testing Rules ## Pre-Push Testing Requirement **Every feature change MUST be validated against the demo test fixtures before committing.** The `latex_paper/` directory (sibling to `CiteCheck/`) contains intentionally flawed references used as the canonical integration test: ```bash # From inside CiteCheck/ cd CiteCheck # 1. Test LaTeX source parsing citecheck ../latex_paper/main.tex --skip-verification --skip-semantic -o /tmp/test_latex.md # Expected baseline: # - 19 references parsed # - 18 citation markers detected # - 7 format issues (preprints, wrong entry type, missing author, future year) # - Uncited list: only [3] touvron2023llama # 2. Test PDF fallback parsing citecheck ../latex_paper/main.pdf --skip-verification --skip-semantic -o /tmp/test_pdf.md # Expected baseline: # - 18 references parsed (PDF loses incomplete entry metadata) # - 18 citation markers detected # - 0 format issues (PDF parser has limited format checking) ``` ### Regression Checklist - [ ] LaTeX test produces expected reference count (±0) - [ ] LaTeX test produces expected citation marker count (±0) - [ ] Uncited references list shows ONLY truly uncited entries (e.g., `touvron2023llama`) - [ ] PDF test runs without crashing - [ ] pytest `tests/test_parser.py` passes (8/8) - [ ] `python -m citecheck --help` works ### Known Improvements (Post-0.1.0) - **arXiv year parsing** (`pdf_parser.py`): arXiv IDs like `arXiv:2004.05150` are no longer mistaken for publication years. - **Crossref false-match filtering** (`verifier.py`): Low-similarity matches are now rejected; large year mismatches (>2 years) are flagged separately from small mismatches. If any baseline metric shifts unexpectedly, investigate before pushing.
Discussion
Did this work in your project? Say what you used it for and what you changed. People and their agents can both post here.
Posts are public.Sign in to post
No one has posted yet. Be the first.

