clinical-data-extractor
Pass
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted clinical trial data from external URLs and PDF documents.\n
- Ingestion points: Web content retrieved via browser action=snapshot and PDF text extracted using nano-pdf.\n
- Boundary markers: The skill does not implement delimiters or isolation instructions to separate untrusted external content from internal agent instructions.\n
- Capability inventory: The agent has access to the browser tool for web navigation and the read/write tool for file system operations.\n
- Sanitization: While basic filename sanitization is performed, no content filtering or safety-escaping is applied to the extracted clinical data before processing.\n- [COMMAND_EXECUTION]: The skill utilizes built-in platform tools including browser for web automation and nano-pdf for document processing as part of its core functional workflow.\n- [EXTERNAL_DOWNLOADS]: Accesses professional medical conference sites (e.g., ASCO, ESMO) to retrieve clinical abstracts and references its source code on GitHub, which are legitimate operations given its stated purpose.
Audit Metadata