competitive-intel

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md directs the agent to perform web search and then fetch and read arbitrary outsider-authored web content (including community sources like Reddit/HN) before producing the report, so an outsider can inject poison text via content that the workflow selects and ingests.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly instructs runtime fetching via the Jina Reader curl command (https://r.jina.ai/https://example.com), which pulls arbitrary remote page content that would be injected into the agent's context and could therefore directly control prompts.