MIND-Skill: Quality-Guaranteed Skill Generation via Multi-Agent Induction and Deduction

Yixuan Li 1 Mingshu Cai 2 Ziyang Xiao 3 Wanyuan Wang 4 Yanchen Deng 1 Bo An 1 1 Nanyang Technological University 2 Waseda University 3 Zhejiang University 4 Southeast University · 2026

Read MIND-Skill: Quality-Guaranteed Skill Generation via Multi-Agent Induction and Deduction by Yixuan Li 1 Mingshu Cai 2 Ziyang Xiao 3 Wanyuan Wang 4 Yanchen Deng 1 Bo An 1 1 Nanyang Technological University 2 Waseda University 3 Zhejiang University 4 Southeast University (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-08670.txt.

Description

Li and five colleagues automate reusable agent skills, the distilled strategies that help AI agents solve multi step tasks. Agents still struggle with complex, multi step real world work needing domain specific procedural knowledge, and hand curation demands scarce experts. Their framework, MIND-Skill, induces general skills from wins with an induction agent and a deduction agent.

The induction agent abstracts reusable skills from wins, while the deduction agent rebuilds trajectories by following those skills. Quality is guarded by three textual losses: reconstruction loss comparing input and rebuilt trajectories, outcome loss enforcing correctness, and rubric loss judging documentation and abstraction level. Those losses are jointly optimised with TextGrad, and the skills are tested on held out tasks unseen during optimisation. On AppWorld and BFCL-v3, MIND-Skill consistently beats concurrent skill generation methods. The work turns skill curation from manual craft into an automated pipeline with quality checks.

Catalog record

CREATORYixuan Li 1 Mingshu Cai 2 Ziyang Xiao 3 Wanyuan Wang 4 Yanchen Deng 1 Bo An 1 1 Nanyang Technological University 2 Waseda University 3 Zhejiang University 4 Southeast University
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE60 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.08670

Browse the full library · About the Tower of Babel