anyterminal_hermes:
  responses_api_agents:
    anyterminal_agent:
      entrypoint: app.py
      domain: coding
      description: Terminal Bench run by the Hermes agent inside the task container.
      value: Evaluate terminal-task capabilities on Terminal Bench with any Gym agent.

      agent_server_module: responses_api_agents.hermes_agent.app
      agent_server_class: HermesAgent
      agent_config_class: HermesAgentConfig
      agent_kwargs:
        max_turns: 90
        terminal_backend: local
        terminal_timeout: 180
        compression_enabled: true
        compression_threshold: 0.85

      # Path to pre-built Apptainer SIF files produced by prepare.py --sif-dir.
      # If omitted, Apptainer will pull docker:// images at runtime (requires internet).
      tb_sif_dir: null

      apptainer_memory_limit_mb: 32768
      concurrency: 6

      model_server:
        name: policy_model
        type: responses_api_models

      datasets:
        - name: example
          type: example
          jsonl_fpath: responses_api_agents/anyterminal_agent/data/terminal_bench.jsonl