The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The installation instructions require cloning a repository from an unverified GitHub user account (https://github.com/fahad-hamid/psique-workflow-clinic.git) and executing npm install. Since this source is not from a known trusted organization, it introduces a risk of executing unvetted code during the setup process.
[INDIRECT_PROMPT_INJECTION]: The ClaudeService in backend/src/ai/claude.service.ts processes clinical notes, patient context, and historical records to generate summaries and risk assessments. This architecture is vulnerable to indirect prompt injection if malicious instructions are embedded within the patient data or clinical notes.
Ingestion points: Untrusted clinical notes and patient history data are passed directly into prompt templates in the summarizeClinicalNotes, generateQuickResponse, and analyzePatientRisk methods.
Boundary markers: The prompts lack clear delimiters or instructions to ignore potential commands embedded within the clinical data.
Capability inventory: The AI outputs are used to generate professional clinical summaries and automated risk assessments, which could be manipulated to produce misleading or harmful therapeutic guidance.
Sanitization: There is no evidence of filtering or sanitization of the input text before it is interpolated into the system prompts.

sesion-mental-health-saas