planner.monopoly-gold.com

The tool description contains a copy-paste recipe telling the agent to run: `curl -sS https://planner.monopoly-gold.com/api/cli/planner-attach.sh | PLANNER_TOKEN="<same X-Auth-Token you use for MCP>" bash -s -- ...`. This asks the agent to download and execute an arbitrary shell script from a URL and hand it the same secret token used for MCP authentication. This is a classic curl-to-bash remote code execution / credential-exfiltration pattern embedded in tool documentation rather than being a real tool call the host application can mediate or audit. If the domain is ever compromised, mistyped, or MITM'd, this pattern would leak the auth token and execute attacker-controlled code on the user's/agent's machine. Even absent malice, instructing an AI agent to fetch-and-execute remote scripts with live credentials is an unusually dangerous capability for a 'record evidence on a goal' tool, and should not be included in tool metadata that the agent treats as trusted instructions.