folk

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

• Third-party content exposure detected (high risk: 0.80). The SKILL.md explicitly instructs the agent to run "membrane action run" and "membrane request" against the Folk API (see "Running actions" and "Proxy requests"), which fetches user-generated third-party contact/record data that the agent is expected to read and act on, so that content could indirectly inject instructions influencing subsequent actions.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.70). The skill explicitly requires fetching and running the Membrane CLI at runtime (via "npm install -g @membranehq/cli" or "npx @membranehq/cli@latest"), which executes remote code from the package (e.g. https://www.npmjs.com/package/@membranehq/cli) and is a required dependency for the skill to operate.