drug-discovery
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches scientific data from reputable domains including ebi.ac.uk, ncbi.nlm.nih.gov, fda.gov, and opentargets.org. These are well-known services for chemical and pharmacological research.\n- [REMOTE_CODE_EXECUTION]: While automated scanners detected pipes to Python, manual analysis confirms these are used to process JSON data from trusted APIs using local scripts. No remote code is executed from external servers.\n- [COMMAND_EXECUTION]: The skill executes shell commands (curl, python) to fetch and process data. User inputs are properly URL-encoded using Python's standard library to mitigate command injection risks.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external API data, creating a potential surface for indirect prompt injection.\n
- Ingestion points: Data from ChEMBL, PubChem, and OpenFDA labels are processed in SKILL.md and accompanying scripts.\n
- Boundary markers: Output is labeled with clear headers but lacks explicit 'ignore instructions' delimiters for the ingested data.\n
- Capability inventory: The skill has network access and shell execution capabilities.\n
- Sanitization: Data is parsed as JSON, which limits the attack surface, but specific string fields are not sanitized for embedded NL instructions.
Audit Metadata