Learning Globally Reusable Skills for Coding Agents

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Chen Yang, Jiashuo Tian, Ziqi Wang, Xinyin Liu, Meiru Ye, Junjie Chen
• Venue: arXiv: cs.SE
• Published: 2026-08-06
• Collected: 2026-08-07
• URL: http://arxiv.org/abs/2608.06153v1
• PDF: https://arxiv.org/pdf/2608.06153v1
• DOI: UNKNOWN

Plain-Language Brief

Researchers present a framework called GSE (Globalized Skill Evolution) that aims to improve automated skill evolution for Large Language Model (LLM) agents in software engineering tasks. The proposed method addresses the limitations of existing approaches by jointly optimizing skill compatibility and generalization, preserving consistency across skills, and preventing overfitting. The authors evaluate GSE on two tasks: bug-revealing test generation and false-positive bug report filtering, showing improved precision, recall, and F1-score compared to existing evolution techniques.

What the Document Says

• Automated skill evolution for LLM agents is discussed.
• GSE framework proposed for globalized skill evolution.
• GSE models inter-skill relationships using a Skill Relation Graph (SRG).
• GSE performs cluster-based skill consolidation and replay-driven verification to improve generalization and prevent overfitting.
• Evaluation of GSE on two software engineering tasks: bug-revealing test generation and false-positive bug report filtering.
• Comparisons show improved precision, recall, and F1-score for GSE compared to existing evolution techniques.
• Improved results also shown when deploying GSE on an internal industrial agent.

Key Concepts Mentioned

• Automated skill evolution
• Large Language Model (LLM) agents
• GSE framework
• Skill Relation Graph (SRG)
• Cluster-based skill consolidation
• Replay-driven verification
• Bug-revealing test generation
• False-positive bug report filtering

Claims Made by the Document

• Existing skill evolution approaches are limited.
• GSE improves precision, recall, and F1-score in software engineering tasks.

Evidence Used

• Comparisons with existing evolution techniques.
• Improved results shown for GSE on various tasks.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Automated skill evolution enables Large Language Model (LLM) agents to continuously improve without expensive retraining. However, existing approaches typically treat skill evolution as a sequence of local updates, overlooking relationships among skills and often producing overfitted skill updates that fail to generalize across tasks. We propose GSE, a globalized skill evolution framework that jointly optimizes skill compatibility and skill generalization. To preserve consistency across the skill bank, GSE maintains a Skill Relation Graph (SRG) that explicitly models and co-evolves inter-skill relationships. To improve generalization, GSE performs cluster-based skill consolidation to abstract reusable capabilities from local updates and employs replay-driven verification to prevent overfitting and behavioral regressions. We evaluate GSE on two representative software engineering tasks: bug-revealing test generation and false-positive bug report filtering. Across two state-of-the-art coding agents, OpenHands and mini-SWE-agent, GSE consistently achieves the best precision, recall, and F1-score. Compared with existing evolution techniques, GSE improves precision and recall by 6.1%~34.1% and 31.8%~180.0% for test generation, and by 15.4%~96.4% and 13.1%~19.8% for false-positive filtering. Deployment on an internal industrial agent further yields a 61.4% improvement in F1-score, demonstrating the effectiveness and generalizability of GSE for evolving effective skills.