| - | Shared | Append the effective non-public rollout network policy to the harness input. |
| - | Shared | Stirrup package version to install. |
| - | Shared | How to prepare the stirrup package in the env. |
| Shared | Maximum agent turns passed to stirrup.Agent. |
| Shared | Maximum output tokens per model response. |
| Shared | Model context window; stirrup summarizes history when usage nears it. |
| Shared | Per-command wall-clock timeout in seconds for the code_exec tool. |
| Shared | Timeout in seconds for one model request. |
| Shared | Total model attempts including the first; native max_retries counts additional retries. |
| - | Harness-specific | Expose Stirrup’s view_image tool. Set to true when the Benchmark’s modality includes image, where an explicit false is an error. Disable for text-only models, which reject image input. |
| Harness-specific | Additional model request parameters; explicit model.params values take precedence. |
| Harness-specific | Fraction of the context window used before summarizing conversation history. |
| Harness-specific | Remaining turns at which the agent starts receiving a limit warning. |
| Harness-specific | Run synchronous tool executors in a worker thread. |
| Harness-specific | Move images from tool results into separate user messages. |
| Harness-specific | Ask the agent to continue when an assistant response contains no tool calls. |
| Harness-specific | Summarize and retry after a rejected context overflow. |