Know Before Fix: QA-Driven Repository Knowledge Acquisition for Software Issue Resolution
AuthorsHaotian Lin, Silin Chen, Xiaodong Gu, Yuling Shi, Chengxi Pan, Jiaqi Ge, Mengfan Li, Jianghong Huang, Mengchieh Chuang, Beijun Shen, Haibing Guan
Resources
ACQUIRE helps coding agents fix software issues more accurately by having them explicitly ask and answer repository-specific questions before writing a patch.
Key results
Number of real GitHub issue instances used for evaluation.
Single-attempt issue-resolution rate on SWE-bench Verified.
Single-attempt issue-resolution rate using OpenAI’s GPT-5-mini.
Audited QA pairs whose central claims were supported by repository evidence.
Reduction in mean agent rounds on instances recovered by QA injection.
Pass@1 decrease when structured QA decomposition is replaced by a single proposal.
What the paper found
The paper introduces ACQUIRE, a QA-driven framework that separates repository understanding from patch generation. Before editing, a Questioner decomposes each issue into targeted questions across Mechanism and Behavior, Design and Usage, Locating and Structure, and Ecosystem and Standards; independent Answerers explore the repository in read-only mode and return evidence-grounded answers, after which a Resolver uses the structured QA context to implement and test a patch. On SWE-bench Verified, a benchmark of 500 real GitHub issues, ACQUIRE reaches 70.8% Pass@1 with DeepSeek-V3.2 and 62.2% with OpenAI’s GPT-5-mini, outperforming the shared Mini-SWE-Agent baseline while adding modest cost and latency. Human review found 99.1% of audited QA pairs supported by repository evidence. The framework reduces agent rounds by 17.1% on instances recovered from failure, concentrating effort on faster localization and fixing while increasing the relative emphasis on reproduction and verification. Ablations show that replacing targeted question decomposition with a single repair proposal lowers Pass@1 by 4.8 percentage points, demonstrating that explicit knowledge-gap interrogation is more effective than asking an agent to jump directly from issue description to solution. The authors conclude that repository-grounded QA improves accuracy, efficiency, interpretability, and cross-model generalization, although misleading repair framings remain a limitation.
Original abstract
LLM-based coding agents have significantly advanced automated software issue resolution, yet they remain highly prone to factual errors caused by insufficient repository understanding. Recent methods attempt to mitigate this limitation through pre-repair repository exploration; however, their fix-driven strategies explore repositories without identifying the agent's knowledge gaps, often yielding imprecise context that fails to bridge the underlying understanding deficit. In this paper, we propose ACQUIRE, a QA-driven framework for software issue resolution. Mirroring how experienced developers first comprehend unfamiliar code before attempting a fix, ACQUIRE explicitly acquires repository knowledge prior to repair. The framework decouples knowledge acquisition from patch generation through two stages: in the first stage, a Questioner and an Answerer collaborate to acquire structured repository knowledge, where the Questioner poses targeted questions and the Answerer produces evidence-grounded answers through autonomous exploration; in the second stage, the Resolver leverages the resulting QA knowledge to generate informed patches. By transforming implicit knowledge gaps into explicit, factually reliable understanding, ACQUIRE accelerates knowledge-intensive repair stages and enables more accurate resolution. 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.
Read the original paperMore in Code Generation
Browse all 43 papers →Compact Documentation for Coding Agents: A Benchmark, an Optimizer, and Why It Does Not Transfer
Md Shohel Arman, Igor Molybog
Better code documentation can faithfully reconstruct software, but surprisingly does not necessarily help AI coding agents fix real issues when the source code is already available.
Groupwise Agentic Grading and Advantage Redistribution for Code Agent RL
Jinhao Dong, Liang Zhao, Zihao Yue, Wenhan Ma, Linghao Zhang, Lei Li, Shicheng Li, Yifan Song, Bowen Ye, Fuli Luo
GAGAR helps code agents learn not only to pass tests, but to produce cleaner and more targeted implementations by redistributing RL credit according to agentic quality judgments.
Reinforcement Learning from Intermediate Renders for Image-to-Code Generation
Omri Kaduri, Kate Feingold, Phillip Isola, Tali Dekel
IR4RL improves image-to-code generation by rewarding models for making useful visual progress at every intermediate rendering step.