validate-bib
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from user-provided bibliography files (
.bib) and document files (.tex,.qmd,.md). Maliciously crafted fields within these files—such as a paper's title, author, or citation key—could contain instructions designed to manipulate the agent's behavior when it later reads the generated audit reports inquality_reports/. - Ingestion points: Project documents (
Slides/*.tex,Quarto/*.qmd,guide/*.qmd,master_supporting_docs/**/*.tex) and the primary bibliography file (Bibliography_base.bib). - Boundary markers: The skill description does not specify the use of delimiters or 'ignore' instructions when interpolating extracted metadata into the markdown reports.
- Capability inventory: The skill uses
Bash,Write,Read, andWebFetchtools, providing a significant surface for subsequent actions if an injection is successful. - Sanitization: There is no mention of sanitizing or escaping extracted strings (like titles or author names) before they are written to the final audit report.
Audit Metadata