exa-contents

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

• Third-party content exposure detected (high risk: 0.85). The skill’s /contents workflow sends caller-supplied urls to Exa, and Exa returns extracted page text/highlights/summary (LLM-readable prose) from those outsider-authored web pages into the agent context, which is an indirect prompt-injection path via arbitrary public web content.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill makes runtime calls to the Exa API (POST https://api.exa.ai/contents) which crawls arbitrary target URLs and returns extracted text/highlights that are intended to be injected into the agent's context, so fetched remote content can directly control prompts.