competitive-intel

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

• Third-party content exposure detected (high risk: 0.95). The skill’s required runtime workflow includes web search and URL fetching (e.g., via MCP DuckDuckGo search/Jina fetch, built-in web tools, or Jina Reader), which ingests outsider-authored free text from arbitrary public web pages 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). I flagged the Jina Reader URL pattern (e.g., https://r.jina.ai/https://example.com) because the skill explicitly instructs using that runtime fetch to load arbitrary webpage content which would be injected into the agent context and thus can directly control prompts.