bgpt-paper-search

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

• Third-party content exposure detected (high risk: 0.85). The skill requires the agent to call a remote MCP tool (search_papers on https://bgpt.pro/mcp/sse), which will return readable paper-derived text/fields from an external third-party database at runtime—an outsider-authored source.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill config explicitly runs remote code via npx (e.g., "npx mcp-remote") and connects at runtime to the MCP SSE endpoint https://bgpt.pro/mcp/sse (and links https://bgpt.pro/mcp), which will execute/fetch remote code and supply tool instructions that directly control the agent's prompts and behavior.