secure-python-code-sandbox

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

• Potentially malicious external URL detected (high risk: 0.90). The skill instructs callers to fetch live schema/instructions via the AgentPMT MCP (e.g., calling get_instructions) which makes runtime requests to the AgentPMT MCP server https://api.agentpmt.com/mcp/ and the returned content is treated as authoritative, so remote content can directly control prompts/instructions.