deep-research

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits a vulnerability surface for indirect prompt injection (Category 8).
  • Ingestion points: The skill ingests untrusted data from an external literature corpus and scans PDF documents using the pdftotext utility (as described in bibliography_agent.md and timeline_extraction_agent.md).
  • Boundary markers: The skill employs structured boundary markers, such as <!--ref:slug--> and <!--anchor:kind:value--> HTML comments, to isolate citation metadata within report drafts (documented in report_compiler_agent.md).
  • Capability inventory: The agents can perform network requests to academic APIs, execute shell commands (specifically pdftotext), and write files to phase-specific directories.
  • Sanitization: The provided instructions do not detail explicit sanitization or filtering logic to prevent the agent from following instructions that may be embedded within the text of the academic papers or PDFs it processes.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 05:06 AM
Security Audit — agent-trust-hub — deep-research