← ALB News · curated headline

Knowledge-Centric Automated Issue Resolution: A Systematic Survey and Taxonomy

arXiv Software Engineering · WORLD · Compilers · 11/17/2026, 12:00:00 AM

arXiv:2611.18569v1 Announce Type: new Abstract: Automated issue resolution is a knowledge-intensive software maintenance task in which LLM-based agents must acquire, represent, use, and update knowledge from repositories, execution environments, historical artifacts, documents, tools, and prior repair processes. This paper presents a systematic survey of 133 studies on automated issue resolution from a knowledge-centric perspective. We introduce a three-layer taxonomy that organizes issue-resolution knowledge into the Background Knowledge Layer, the Repository Knowledge Layer, and the Procedural Knowledge Layer, and analyze the knowledge lifecycle along six dimensions: knowledge types, sources and extraction methods, representation formats, usage methods, workflow stages, and temporal evolution. Our analysis shows that current systems primarily rely on Existing Code Knowledge and Development Tool Knowledge, while Repository Evolution Knowledge and Experiential Knowledge are increasingly adopted, indicating a shift toward history-informed, feedback-aware, and experience-oriented issue resolution. In contrast, Design Architecture Knowledge and explicitly modeled Background Knowledge remain comparatively underexplored. These findings suggest that future systems should move beyond temporary context construction toward reusable, traceable, and evolvable software knowledge infrastructures that support more controllable, verifiable, and generalizable software engineering agents. The surveyed studies and related resources are maintained at https://github.com/SYSUSELab/Awesome-Knowledge-for-AIR.

Continue to arXiv Software Engineering →

You are being forwarded in 6s — stay on this page