web-search-linkup

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

• Third-party content exposure detected (high risk: 0.95). The skill’s REQUIRED workflow includes calling the runtime /search and/or /fetch endpoints with user-supplied queries/URLs, which causes the system to ingest outsider-authored web page text (public web content) into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill calls the Gooseworks proxy API (default $GOOSEWORKS_API_BASE -> https://api.gooseworks.ai via endpoints like /v1/proxy/orthogonal/run and /v1/proxy/orthogonal/fetch) to retrieve arbitrary webpages (e.g., "https://example.com/article") and exposes a renderJs option that executes page JavaScript and returns raw HTML for injection into agent outputs, so runtime-fetched external content can directly execute code or control prompts.