The Agent Skills Directory

Indirect Prompt Injection (SAFE): The skill processes user-provided LaTeX and BibTeX files through scripts like extract_prose.py and verify_bib.py.
Ingestion points: LaTeX (.tex), Typst (.typ), and BibTeX (.bib) files are read and parsed.
Boundary markers: None explicitly implemented to separate data from potential instructions within the LaTeX source.
Capability inventory: Limited to executing chktex via subprocess and extracting text for analysis.
Sanitization: No explicit sanitization of document content is performed; however, the primary consumer is the user/agent for writing assistance, posing minimal risk.
Unverifiable Dependencies (SAFE): The check_figures.py script requires the Pillow (PIL) library for image analysis. This is a standard, reputable library for the stated purpose.
Command Execution (SAFE): The check_format.py script uses subprocess.run to call the chktex linter. The execution is restricted to the specific binary and the user-specified file path, following best practices for local tool integration.

latex-paper-en