deep-research

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

• Third-party content exposure detected (high risk: 0.95). The skill’s runtime workflow performs public web search and full-page fetching (via web-forager search/fetch or ddgs + r.jina.ai), so outsider-authored free text from arbitrary fetched URLs is ingested into the agent’s LLM context during synthesis.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill fetches arbitrary web pages at runtime (e.g., via "uvx ... web-forager fetch "https://example.com\"" and the Jina Reader proxy "https://r.jina.ai/https://example.com"), and that fetched content is injected into the agent's context to drive prompts/synthesis, so it can directly control agent behavior.