arXiv cs.AI· Zekai Li, Zhenke Liu, Yonghao Li, Junhan Liu, Xulong Xiao, Jiapeng Zhang·· 12 小时前AI 评分34
LIBRA：通过运行时组装自动归纳可执行工具库
LIBRA: Programmatic Induction of Executable Tool Libraries
AI 导读
LIBRA 将扁平工具目录转为可执行工具库，由 LLM 提出每个工具的调用契约、可复用输出、状态变更与成功条件，并在沙箱中验证组合关系后保留运行时成立的部分。该工具库在用户查询前构建，无需依赖标注、工具使用轨迹或任务特定训练，支持一次性与智能体式检索。
正文 · 原文
Abstract:Large language model agents increasingly retrieve tools from large catalogs, yet most retrievers rank each tool independently by request similarity. For multi tool tasks, this recovers the goal matching tool while omitting prerequisite tools required by the this http URL introduce LIBRA, a framework for Library Induction by Runtime Assembly that turns a flat tool catalog into an executable tool library. An LLM proposes per tool contracts of candidate invocations, reusable outputs and state changes, and success conditions; LIBRA then tests these compositions in a sandbox and retains the relations supported at runtime. Constructed before user queries, the library requires no dependency annotations, tool use trajectories, or task specific training, and supports both one-shot and agentic retrieval. On AppWorld, BFCL, and SciToolEval, LIBRA improves one-shot Recall and NDCG over the same dense retriever by 7.9 and 3.7 percentage points on average. Within ToolReAGt, it yields corresponding gains of 4.5 and 2.9 points after ten retrieval iterations.
| Comments: | 10 pages, 5 figures. Introduces LIBRA, a query-agnostic tool library built via execution-validated prerequisites in a sandbox (avg ARR review score: 3.5/5). Demonstrates consistent performance gains across three benchmarks while discussing sandbox dependencies and library construction trade-offs |
| Subjects: | Software Engineering (cs.SE); Artificial Intelligence (cs.AI); Information Retrieval (cs.IR) |
| Cite as: | arXiv:2611.05445 [cs.SE] |
| (or arXiv:2611.05445v1 [cs.SE] for this version) | |
| https://doi.org/10.48550/arXiv.2611.05445 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Zhenke Liu [view email]
[v1]
Tue, 3 Nov 2026 16:57:04 UTC (1,672 KB)
来源：arXiv cs.AI · arxiv.org