anyterminal_claude_code: responses_api_agents: anyterminal_agent: entrypoint: app.py domain: coding description: Terminal Bench run by claude-code natively inside the task container. value: Evaluate terminal-task capabilities on Terminal Bench with any Gym agent. agent_server_module: responses_api_agents.claude_code_agent.app agent_server_class: ClaudeCodeAgent agent_config_class: ClaudeCodeAgentConfig agent_kwargs: model: ${policy_model_name} anthropic_api_key: ${policy_api_key} anthropic_base_url: ${policy_base_url} timeout: 12500 # Max task timeout in Terminal Bench is 12000 seconds max_turns: null # Unlimited turns container_formatter: "docker://{docker_image}" concurrency: 6 # Claude Code calls anthropic_base_url directly (no Gym model server). datasets: - name: example type: example jsonl_fpath: responses_api_agents/anyterminal_agent/data/terminal_bench.jsonl anyterminal_claude_code_environment_server: environment_servers: legacy_agent: entrypoint: app.py agent_server: type: responses_api_agents name: anyterminal_claude_code