agent-survey-corpus
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS] (LOW): Downloads PDFs from arXiv. While arXiv is a reputable academic repository, any network-based content ingestion presents a potential risk surface.
- [COMMAND_EXECUTION] (LOW): Invokes a local Python script (
run.py) to process the downloads. This is common for skill functionality. - [PROMPT_INJECTION] (LOW): Potential for Indirect Prompt Injection. The skill extracts text from external documents which are then used as stylistic references for the agent. Malicious documents could contain hidden instructions intended to influence the agent's behavior. Evidence Chain: (1) Ingestion points: arxiv_ids.txt and downloaded arXiv PDFs. (2) Boundary markers: None provided to separate instructions from extracted text. (3) Capability inventory: Local script execution and file system writes. (4) Sanitization: None mentioned.
Audit Metadata