sesion-mental-health-clinic-platform

Warn

Audited by Gen Agent Trust Hub on Jul 13, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user to download the primary application source code from a third-party GitHub repository (github.com/fahad-hamid/psique-workflow-clinic). This repository is not associated with a verified organization or the listed skill author.
  • [REMOTE_CODE_EXECUTION]: After downloading the external source code, the instructions direct the user to run pnpm install and pnpm dev. This executes third-party code locally, presenting a supply chain risk if the repository content is untrusted.
  • [COMMAND_EXECUTION]: The setup process involves running several shell commands, including git clone, docker-compose, and pnpm, to provision the environment and application services.
  • [PROMPT_INJECTION]: The application possesses an indirect prompt injection surface in the ClaudeService. Untrusted clinical notes and session transcripts are interpolated directly into AI system prompts for summarization and risk detection.
  • Ingestion points: Clinical notes and session transcripts in ClaudeService (summarizeSession, detectRiskFactors).
  • Boundary markers: Absent; the session content is concatenated directly into the prompt string without delimiters.
  • Capability inventory: Summarization, evolution report generation, and risk factor detection using Anthropic Claude models.
  • Sanitization: No visible sanitization or filtering of the clinical session content is implemented before AI processing.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 13, 2026, 06:06 AM
Security Audit — agent-trust-hub — sesion-mental-health-clinic-platform