sesion-clinic-workflow
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The
ClaudeServicelocated inapps/backend/src/ai/claude.service.tsprocesses untrusted clinical notes for summarization and drafting assistance by interpolating them directly into LLM prompts. - Ingestion points: The skill retrieves patient clinical notes from the database and passes them to the AI orchestration methods.
- Boundary markers: The prompts utilize basic headers such as 'Notas de sesión:' to delimit content but lack robust delimiters or specific instructions for the model to disregard potentially malicious commands embedded within the notes.
- Capability inventory: Generated summaries are written back to the database (
aiSummary), which could influence downstream clinical decision-making or automated reports. - Sanitization: There is no evidence of input sanitization or escaping of the clinical text prior to its inclusion in the prompt.
- [EXTERNAL_DOWNLOADS]: The skill downloads project source code from an external GitHub repository (
https://github.com/fahad-hamid/psique-workflow-clinic.git) to initialize the application environment. This is a standard procedure for this type of developer workflow tool.
Audit Metadata