research

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

• Third-party content exposure detected (high risk: 0.95). Outsider free text is ingested when the user runs /research <url> and the workflow calls WebFetch(url, ...) (or Chrome DevTools take_snapshot) to extract arbitrary third-party article text, which is then used to generate and write KB insight content into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill fetches arbitrary HTTP/HTTPS article URLs at runtime (via WebFetch(url, "...") and the Chrome DevTools MCP fallback) and injects the fetched article text into the agent's processing/extraction prompts, so the provided article URL is a runtime external dependency that can directly control the agent.