web-search

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

• Third-party content exposure detected (high risk: 0.95). When --fetch is used, the skill fetches arbitrary public web pages at runtime (fetchPageContent() → fetch(url) → response.text() → article.text()/body.text()), which is outsider-authored free text that can include prompt-injection content.

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

• Potentially malicious external URL detected (high risk: 0.70). This skill performs runtime fetches from https://duckduckgo.com/html/?q=... and (with --fetch) retrieves arbitrary result page URLs (the decoded hrefs) and returns their page content into the tool output, which can be injected into the agent/model context and therefore directly influence prompts/responses.