github-actions

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

• Third-party content exposure detected (high risk: 0.70). This skill connects to GitHub Actions via the Membrane connector and explicitly instructs the agent to list and run actions (e.g., "membrane action list --connectionId=..." and "membrane action run "), which means it fetches and interprets user-generated GitHub workflow/action metadata (third‑party content) that can influence tool decisions and execution.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill requires fetching and running the remote Membrane CLI (e.g., "npm install -g @membranehq/cli@latest" and "npx @membranehq/cli@latest"), which downloads and executes third-party code at runtime and is described as a required dependency for the skill.