Claude Managed Agents permission policies now include auto

Anthropic's Claude Managed Agents permission policies now include an "auto" option, alongside a new terminal command to act on it.

• Under auto, the server evaluates each agent or MCP tool call itself and runs it, denies it, or pauses for approval, rather than asking on every call. The agent.tool_use and agent.mcp_tool_use events report how each call was evaluated in a new evaluation field alongside evaluated_permission.
• The ant CLI adds ant beta:sessions connect, which attaches a terminal to a running Managed Agents session, loads its transcript, and follows it live. From there you can send a message, interrupt the agent, or allow or deny a tool call waiting for approval, whether because of an always_ask policy or because auto reached no determination. Passing --web opens the session in the Claude Console's session viewer in a browser instead, where every thread of a multiagent session is visible, not just the primary one.

Anthropic

Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.

View Anthropic →