We study whether LLMs can produce executable workflow artifacts that remain consistent across graph structure, tool implementation, schema bindings, and runt…
机构:上交
来源:arXiv 2609.23925 | AI4Papers 论文推荐平台