The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the agent to clone a source code repository from an unverified third-party GitHub user (fahad-hamid/psique-workflow-clinic), which conflicts with the 'ara.so' (Aradotso) authorship claimed in the metadata.
[REMOTE_CODE_EXECUTION]: By instructing the agent to run pnpm install and development servers on the cloned repository, the skill enables the execution of arbitrary code defined in the external repository's configuration files.
[COMMAND_EXECUTION]: The installation and setup process requires several privileged shell commands, including git clone, docker run for service orchestration, and pnpm prisma for database schema modifications.
[CREDENTIALS_UNSAFE]: The configuration documentation requires the management of sensitive secrets and file-based credentials, including AFIP electronic billing certificates (.pem and .key files), and private API keys for Anthropic, Stripe, and Mercado Pago.
[PROMPT_INJECTION]: The skill handles incoming WhatsApp messages for crisis detection and appointment confirmation, creating a surface for indirect prompt injection where untrusted external messages could influence agent behavior. Furthermore, the skill contains metadata poisoning in the form of deceptive references to non-existent AI models ('Claude Opus 4.6' and 'Sonnet 4.6').

sesion-clinic-workflow