CooperBench/team-coop
收藏资源简介:
该数据集包含两个完整的CooperBench团队运行(lead和member共享Redis支持的任务列表、草稿卷和MCP),这些运行被重塑为2代理合作布局,定义在cooperbench/CooperData PR #98中。数据集源自CooperBench的team-trajectories数据集,使用codex代理框架(Azure提供者,非JSON线)和gpt-5.5-hao模型(Azure部署)。总共有1,304个轨迹,全部来自CooperBench的652对完整数据集(每个任务的12个仓库特征对的笛卡尔积),包括通过率信息:一个运行(无协议动词)的通过率为61.8%(403/652),另一个运行(完整团队功能集)的通过率为59.8%(390/652)。数据布局包括代理轨迹、对话、结果和评估文件,映射决策将lead映射为agent1,member映射为agent2,并包含团队特定的额外信息块。
Two complete cooperbench team-harness runs (lead + member sharing a Redis-backed task list, scratchpad volume, and MCP) reshaped into the 2-agent coop layout defined in cooperbench/CooperData PR #98. The dataset is derived from CooperBenchs team-trajectories dataset, using the codex agent framework (Azure provider, non-JSON wire) and model gpt-5.5-hao (Azure deployment). Total of 1,304 trajectories, all from the cooperbench 652-pair full dataset (the 12-repo cartesian product of feature pairs per task), with pass rates: one run (without protocol verbs) at 61.8% (403/652) and another run (full team feature set) at 59.8% (390/652). The layout includes agent trajectories, conversation, result, and evaluation files, with mapping decisions assigning lead to agent1 and member to agent2, and includes a team-specific provenance block.




