dsh-eval¶
⭐ 0 · ✅ active · workflow
One-liner¶
Agent evaluation platform: benchmark YAML, headless dsh runs, trace-based metrics, scripted grading and run comparison.
About¶
Agent Evaluation Platform for deepseek-harness. Run benchmarks against headless dsh profiles, harvest persisted session logs as traces, fold automatic metrics, grade task success and tool selection, and report or compare runs — one benchmark.yaml in, one JSON run + Markdown report out.