The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions require cloning and executing a full microservice stack from an external personal repository: https://github.com/fahad-hamid/psique-workflow-clinic.git.
Evidence: The Installation and Setup section directs the user to git clone and then run pnpm install and pnpm run start:dev on the downloaded content.
[CREDENTIALS_UNSAFE]: The skill defines logic to read highly sensitive tax authority credentials from the local file system.
Evidence: AfipService in backend/src/billing/afip.service.ts uses fs.readFileSync to access AFIP_CERT_PATH and AFIP_KEY_PATH (private keys).
[INDIRECT_PROMPT_INJECTION]: The platform contains an ingestion surface for untrusted external data that could influence agent or system behavior.
Ingestion points: WhatsAppService.handleIncomingMessage in SKILL.md processes incoming WhatsApp messages via the baileys library.
Boundary markers: None present; the content is processed directly from the WAMessage object.
Capability inventory: The skill can emit events, send messages, generate electronic invoices via AFIP, and interact with the Claude AI API.
Sanitization: There is no sanitization or escaping of the message content beyond a simple check for crisis-related keywords.
[COMMAND_EXECUTION]: The setup instructions include multiple shell commands that interact with the system environment and docker infrastructure.
Evidence: The installation guide utilizes docker-compose up, pnpm prisma migrate, and direct execution of backend/frontend services.

sesion-clinic-workflow-platform