# Terminal-Bench 2.1 release 6 contains the complete 89-task suite.
# Keep this release fixed so baseline and escalation runs use identical tasks.
jobs_dir: .nanocodex/harbor/jobs

agents:
  - import_path: harbor_adapter.agent:NanocodexAgent
    model_name: openai/gpt-5.6-sol
    env:
      OPENAI_API_KEY: ${OPENAI_API_KEY}
    kwargs:
      effort: low
      # Match the existing Terminal-Bench policy: built-in local Code Mode
      # tools only. Subagent tools are introduced only in escalation runs.
      web_search: false

datasets:
  - name: terminal-bench/terminal-bench-2-1
    ref: "6"

environment:
  import_path: harbor_adapter.environment:FastDockerEnvironment

verifier:
  import_path: harbor_adapter.verifier:PytestVerifier
