twitter-longform-medical
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [Dynamic Execution] (LOW): The file
references/visualization-guide.mdcontains multiple Python code blocks intended for generating charts. These templates usematplotlibto perform file system writes (plt.savefig) to export visualizations as PNG files. This is a legitimate functional requirement for data reporting but involves the execution of script-based logic. - [Indirect Prompt Injection] (LOW): The skill is designed to process external clinical data (e.g., trial results). This creates a vulnerability surface where untrusted data ingested by the agent could contain malicious instructions. Ingestion point: External trial data described in
references/voice-guide.md. Boundary markers: Absent. Capability inventory: Script execution for visualization and data formatting. Sanitization: Not addressed in the reference files.
Audit Metadata