agents-sdk

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md explicitly instructs the agent/operator to inspect and read external public sources (e.g., "Inspect the target repo. Read README.md..." and to clone or pull the public cookbook at https://github.com/openai/openai-cookbook and read official docs URLs), meaning untrusted third-party web/GitHub content will be fetched and used to drive build/deploy decisions.

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

• Potentially malicious external URL detected (high risk: 0.90). The Deploy Workflow explicitly instructs cloning and using the openai-cookbook (https://github.com/openai/openai-cookbook), which would be fetched at runtime and then executed via the manager's make/deploy steps, so remote code from that URL would directly be run and is treated as a required dependency.