alterlab-bgpt-search

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

• Third-party content exposure detected (high risk: 0.80). SKILL.md directs the agent to query the remote BGPT MCP server (https://bgpt.pro/mcp/sse / bgpt.pro/mcp) to retrieve structured data extracted from third‑party scientific full‑text papers (methods, results, conclusions), which the agent is expected to read and use to drive literature‑review decisions — creating a clear channel for untrusted external content to influence actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill requires running the MCP client via "npx mcp-remote https://bgpt.pro/mcp/sse" at runtime, which downloads and executes remote code (mcp-remote) and connects to the external MCP endpoint https://bgpt.pro/mcp/sse that can directly control agent behavior.