slurm-cluster-mcp

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

• Third-party content exposure detected (high risk: 0.75). The runtime path uses the MCP client to fetch tool results from an external MCP server over HTTP (scripts/mcp_client.py:55-58 returning result.content[0].text), which can contain outsider-authored free text that the agent would ingest as the tool response.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill's client calls a remote MCP server at runtime (e.g., http://host.openshell.internal:9000/mcp and http://host.docker.internal:8000/mcp), and the server's returned text directly drives the agent's responses, so the external URL(s) control prompts at runtime.