agentic-workflow-benchmark

【数据与研究】Benchmark the efficiency of an agent skill's execution episode in a Copilot session against four goals - minimize LLM token usage, minimize execution time, minimize semantic (model-judgment) steps, and maximize deterministic (script/code-executed) steps - by parsing the session's local events.jsonl telemetry, then persist a structured report with bas…

独立 GitHub 快照 · 自动分类 · 以原作者文档为准