chatgpt-web-research

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly loads the runtime routes "chrome:control-chrome" and "computer-use:computer-use" from the environment skill registry during execution, and those required external skill modules provide the browser-control logic and instructions that directly determine the agent's actions (i.e., they can control prompts/behavior at runtime).