web-search-valyu

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow uses Valyu endpoints like /v1/search, /v1/answer, and /v1/contents that fetch and extract readable web page content from user-supplied URLs/domains, which is outsider-authored free text that can be injected into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill exposes endpoints that fetch and inject arbitrary user-supplied URLs at runtime (e.g., the /v1/contents call with urls like "https://example.com/article"), which will be extracted and included in the AI context/output and thus can directly control prompts.