nlweb-chatgpt-appsdk

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

• Third-party content exposure detected (medium risk: 0.65). The required runtime workflow for the Node.js MCP server path (registering nlweb-list and calling the backend NLWeb /ask endpoint, then returning tool output that the Apps SDK widget renders) can ingest outsider-authored free text if NLWeb /ask fetches or scrapes public web pages or other third-party content at runtime and passes that readable text into the tool result envelope consumed by the LLM/widget.

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

• Potentially malicious external URL detected (high risk: 0.80). The ChatGPT Apps SDK runtime loads and executes the bundled React widget served at ui://widget/nlweb-list.html (served by the Node.js MCP server) during skill runtime, meaning remote HTML/JS is fetched and executed as a required dependency (ui://widget/nlweb-list.html).