deep-research
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted third-party material from web pages, fetched PDFs, and pasted text. It includes a comprehensive 'instruction-data-boundary' pattern across all agent instructions to ensure that imperative-looking text in data is never promoted to a command, thus mitigating injection attempts from external sources.
- [COMMAND_EXECUTION]: The agents are instructed to execute local validation scripts and utility tools such as
scripts/run_ledger.py,scripts/pdf_read_preflight.py, andpdftotext. These commands are strictly defined within the local environment for state tracking and data extraction. - [EXTERNAL_DOWNLOADS]: The skill performs network operations to verify citations against well-known academic services, including arXiv, Crossref, OpenAlex, Semantic Scholar, and PubMed. These are established, trusted platforms for research metadata.
Audit Metadata