cnil-cookie-banner

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

• Third-party content exposure detected (high risk: 0.70). The workflow uses Selenium to load the target website URL at runtime and then reads the page’s visible text/DOM (e.g., self.driver.get(self.base_url) and body_text = self.driver.find_element(..., "body").text), which can include outsider-authored free text from the fetched site and thus be ingested into the agent’s LLM context via the tool output.

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

• Potentially malicious external URL detected (high risk: 0.80). The Python script calls webdriver_manager.chrome.ChromeDriverManager().install(), which at runtime downloads and installs the ChromeDriver binary from external hosts (e.g. https://chromedriver.storage.googleapis.com) and executes that binary as a required dependency, so the skill fetches and runs remote code during execution.