Know Before Fix: QA-Driven Repository Knowledge Acquisition for Software Issue Resolution

2026-03-26 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

ACQUIRE is a QA-driven framework designed to enhance automated software issue resolution by LLM-based coding agents, specifically addressing factual errors caused by insufficient repository understanding. It decouples knowledge acquisition from patch generation into two distinct stages. First, a "Questioner" and an "Answerer" collaborate to acquire structured repository knowledge by posing targeted questions and producing evidence-grounded answers through autonomous exploration. Second, a "Resolver" leverages this acquired QA knowledge to generate informed patches. Experiments on SWE-bench Verified demonstrate that ACQUIRE consistently outperforms representative pre-repair methods, raising Pass@1 by up to 4.4 percentage points with modest additional cost and time, utilizing backbone LLMs like DeepSeek-V3.2 and GPT-5-mini.

Key takeaway

For AI Engineers developing LLM-based coding agents, integrating a QA-driven knowledge acquisition stage like ACQUIRE can significantly enhance resolution accuracy and efficiency. You should prioritize explicit repository understanding before patch generation, leveraging structured question-answering to bridge knowledge gaps. This approach reduces costly trial-and-error, leading to more reliable and interpretable automated bug fixes.

Key insights

Proactive, QA-driven knowledge acquisition significantly improves LLM-based software issue resolution by addressing factual understanding gaps.

Principles

• Decouple knowledge acquisition from patch generation.
• Decompose complex issues into targeted questions.
• Guide question generation with a category taxonomy.

Method

A Questioner poses targeted questions, an Answerer explores the repository to provide evidence-grounded answers, and a Resolver uses this QA knowledge to generate informed patches.

In practice

• Implement a two-stage agentic workflow for bug fixing.
• Use category-guided prompts for knowledge elicitation.
• Prioritize explicit knowledge acquisition before code edits.

Topics

• LLM Agents
• Software Issue Resolution
• Repository QA
• Automated Debugging
• Knowledge Acquisition
• SWE-bench Verified

Code references

• LionLin2003/ACQUIRE
• SWE-agent/mini-swe-agent

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• Automated Root-Cause Subclassification and No-Code Fix Generation for Invalid Bug Reports — Automating invalid bug report subclassification and no-code fix generation significantly reduces software maintenance overhead.
• How to Build a Powerful LLM Knowledge Base — LLM-powered knowledge bases automate information capture and retrieval, significantly enhancing decision-making and operational efficiency.
• Knowledge-Based Pull Requests: A Trusted Workflow for Agent-Mediated Knowledge Collaboration — KPR separates knowledge acceptance from code merging by using agents to distill external contributions into project-controlled implementations.
• Adaptive Intellect Unleashed: The Feasibility of Knowledge Transfer in Large Language Models — Knowledge transfer can improve LLM generalization in software engineering tasks, but its implementation requires careful design.
• Navigating Unreliable Parametric and Contextual Knowledge: Explicit Knowledge Conflict Resolution for LLM Inference — MACR resolves LLM knowledge conflicts by explicitly assessing confidence in sources and employing multi-agent reasoning, moving beyond binary reliability assumptions.
• Diagnosing Knowledge Gaps in LLM Tool Use: An Agentic Benchmark for Novel API Acquisition — LLM tool use with novel APIs requires specific knowledge components, with usage examples being critical, and retrieval/tuning playing complementary roles.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.