hermes-browser-extension

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

• Third-party content exposure detected (high risk: 0.95). At runtime, the extension’s content script scrapes the active web page’s readable text/DOM (e.g., document.body.innerText, headings, links) and sends it as “context” to the Hermes API/LLM, so outsider-authored free text from arbitrary visited websites is ingested into the agent prompt.

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

• Potentially malicious external URL detected (high risk: 0.80). The extension connects at runtime to a Hermes API server (e.g., http://127.0.0.1:8642 or a remote URL such as https://hermes.example.com) and explicitly syncs models, profiles, and skills from that server — external content which directly controls prompts/agent behavior and is required for the extension to function.