anyswe_opencode: responses_api_agents: anyswe_agent: entrypoint: app.py domain: coding description: SWE-bench run by OpenCode inside the task container. value: Eval software engineering capabilities on SWE-bench with any Gym agent. agent_server_module: responses_api_agents.opencode_agent.app agent_server_class: OpenCodeAgent agent_config_class: OpenCodeAgentConfig agent_kwargs: model: model openai_api_key: EMPTY # pragma: allowlist secret workspace_root: /tmp/opencode repo_dir: /testbed thinking: true timeout: 3600 setup_timeout: 300 extra_args: - --dangerously-skip-permissions opencode_version: 1.17.8 system_prompt: Fix the bug in the repository at /testbed. Inspect and edit the source, run relevant checks, and leave the final fix applied. opencode_config: permission: bash: allow edit: allow webfetch: allow provider: nemo: npm: "@ai-sdk/openai-compatible" name: Local model server models: model: name: model interleaved: field: reasoning limit: context: 262144 output: 131072 container_formatter: ${oc.env:ANYSWE_CONTAINER_FORMATTER} sandbox_provider: sandbox sandbox_model_base_url: ${oc.env:NEMO_GYM_SANDBOX_MODEL_BASE_URL,null} swebench_agent_timeout: 3900 swebench_tests_timeout: 1800 concurrency: 4 model_server: name: policy_model type: responses_api_models datasets: - name: example type: example jsonl_fpath: responses_api_agents/anyswe_agent/data/example.jsonl anyswe_opencode_environment_server: environment_servers: legacy_agent: entrypoint: app.py agent_server: type: responses_api_agents name: anyswe_opencode