sesion-mental-health-clinic-platform

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). The required workflow that feeds the LLM is the Claude integration, where runtime patient/session text (e.g., WhatsApp incoming message text and session transcripts/notes) is inserted into the LLM prompt via ClaudeService methods like detectRiskFactors(sessionTranscript) and summarizeSession(sessionNotes, ...), and that text originates from outsiders (patients/WhatsApp contacts) rather than the operating user.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill explicitly includes payment gateway integrations (MercadoPago and Stripe) in the architecture and environment configuration (API keys/secrets present). It also contains an AFIP electronic invoicing service that programmatically authenticates and requests invoices via SOAP (creates and submits invoices). These are specific, finance-related integrations (payment gateways and billing) capable of executing financial/billing operations.