deep-research

Pass

Audited by Gen Agent Trust Hub on Jul 31, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The timeline_extraction_agent is instructed to use the pdftotext utility for extracting metadata from local PDF files. This is a targeted use of a common tool for research purposes.
  • [EXTERNAL_DOWNLOADS]: The pipeline performs automated lookups against reputable academic services for citation verification and metadata retrieval.
  • Evidence:
  • Calls https://api.crossref.org for DOI resolution.
  • Calls https://api.openalex.org for bibliographic triangulation.
  • Calls https://api.semanticscholar.org for graph lookups.
  • Accesses arXiv via http://export.arxiv.org for preprint verification.
  • [PROMPT_INJECTION]: The skill explicitly defines an instruction-data-boundary pattern to ensure content retrieved from external sources is treated strictly as data, effectively mitigating indirect prompt injection risks.
  • [DATA_EXPOSURE]: The skill reads from a structured 'Material Passport' and local research files to maintain project state. It does not attempt to access sensitive system directories, environment variables containing secrets, or user credentials.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 31, 2026, 07:13 AM
Security Audit — agent-trust-hub — deep-research