reward-sciences

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

• Potentially malicious external URL detected (high risk: 0.80). I flag the Membrane CLI fetched/executed at runtime via the npm/npx package @membranehq/cli (e.g., npm install -g @membranehq/cli@latest and npx @membranehq/cli@latest — https://www.npmjs.com/package/@membranehq/cli) because the skill requires installing/running this remote package during operation, which pulls and executes remote code that the agent depends on.