skills/edwinhu/workflows/bib-manage/Gen Agent Trust Hub

bib-manage

Warn

Audited by Gen Agent Trust Hub on Oct 5, 2026

Risk Level: MEDIUMDYNAMIC_CONTEXT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file includes a shell command within the dynamic context syntax (!...) that executes at skill load time. This command attempts to locate and run skill-toc to generate a table of contents for the skill's reference materials.
  • [DYNAMIC_EXECUTION]: The scripts/bibman.py script dynamically loads a local dependency, pincite.py, using importlib.util.spec_from_file_location. This allows the skill to execute code from a sibling project at runtime.
  • [COMMAND_EXECUTION]: The skill uses subprocess.run to execute the local paperpile CLI tool, passing user-derived search queries to retrieve bibliographic data in JSON format.
  • [EXTERNAL_DOWNLOADS]: The skill makes network requests to api.crossref.org and www.govinfo.gov to retrieve bibliographic metadata and Federal Register PDF documents.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted bibliographic data and PDF filenames which are interpolated into search queries and metadata fields.
    • Ingestion points: Bibliographic entries in .bib files and filenames within the project's PDF directory.
    • Boundary markers: None are present to distinguish between trusted instructions and external data in the processed fields.
    • Capability inventory: The skill possesses network access (urllib), shell execution capabilities (subprocess), and the ability to overwrite local bibliography files.
    • Sanitization: The script uses regular expressions and similarity matching to validate data, but does not perform comprehensive sanitization or escaping of the metadata before using it in API queries.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 5, 2026, 10:36 AM