The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted clinical data through AI-assisted services, creating a surface for indirect prompt injection attacks.
Ingestion points: Untrusted clinical notes and patient history data are ingested in packages/ai/src/claude.service.ts through the summarizeSession, chatAssistant, and predictOutcome methods.
Boundary markers: The prompts for these services lack robust boundary delimiters or instructions for the AI to ignore embedded commands, relying instead on simple text headers and newlines.
Capability inventory: The platform has the capability to write to a PostgreSQL database via Prisma (packages/db/prisma/schema.prisma) and send automated WhatsApp messages (packages/whatsapp/src/whatsapp.service.ts) based on processed data.
Sanitization: There is no evidence of input validation, sanitization, or escaping of the clinical text before it is interpolated into system prompts.
[EXTERNAL_DOWNLOADS]: The skill instructions specify the installation of software from an external GitHub repository that does not belong to the identified vendor or trusted organization list.
Evidence: The installation guide in SKILL.md requires running git clone https://github.com/fahad-hamid/psique-workflow-clinic.git.

sesion-mental-health-platform