Workflow-Localized Mechanism Learning: Attribution-Guided Repair and Knowledge Reuse for Structured Agent Skills

Researchers proposed Workflow-Localized Mechanism Learning (WML) to repair and reuse structured agent skills.

WML localizes failing workflow nodes, attributes the responsible mechanism, and applies guided fixes using external knowledge.

Experiments on SpreadsheetBench, WikiTableQuestions, and Compiler-Supported50 show strong performance gains.

The approach improves accuracy, transfer, and cost-efficiency for frozen language-model agents compared with a direct SkillAgent.