researching-web

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md "Reference Following (Deep Research)" section explicitly tells the agent to "Review all cited URLs" and to call WebFetch on cited external URLs (e.g., WebFetch(url="")), meaning the agent fetches and interprets untrusted public web content that can influence its recommendations.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly uses runtime WebFetch calls such as WebFetch(url="", prompt="Extract key details about ") to fetch external pages cited by Perplexity and inject their extracted content into the agent's prompts/context, meaning externally hosted URLs can directly control the agent's behavior.