tooluniverse-drug-research
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests textual data from external scientific databases including PubMed, ChEMBL, and ClinicalTrials.gov. While these are trusted sources, the ingestion of untrusted content such as paper abstracts or trial descriptions constitutes a surface for indirect prompt injection.
- Ingestion points: API responses from PubChem, ChEMBL, PubMed, and FDA FAERS as described in the research paths within
SKILL.mdandREPORT_GUIDELINES.md. - Boundary markers: The skill utilizes a structured markdown template (
REPORT_TEMPLATE.md) to organize data but does not include explicit instructions for the agent to escape or sanitize external inputs. - Capability inventory: The agent has access to over 50 research-focused tool calls and the capability to create and modify local markdown files.
- Sanitization: No explicit sanitization or filtering of external text content is mentioned in the instructions.
- [COMMAND_EXECUTION]: The skill workflow includes a step to execute a local Python script (
pharmacology_ref.py) located in a sibling directory (skills/tooluniverse-drug-drug-interaction/scripts/) to assess drug-drug interactions. This is a functional requirement for the skill's comparative analysis and operates on local files within the skill ecosystem.
Audit Metadata