ACQUIRE: QA Before Repo Patching
SJTU · hf · 2026-07-15
This paper introduces ACQUIRE, aiming to solve a common issue faced by LLM coding agents when fixing software issues: factual errors caused by an insufficient understanding of the codebase.
Core Concept
- Existing pre-fix exploration methods often just "search for context in the repo" without first clarifying what knowledge the agent is actually missing.
- ACQUIRE splits the process into two steps:
- Knowledge Acquisition: A Questioner and an Answerer autonomously explore the repo to raise target questions regarding the issue and provide evidence-backed answers.
- Patch Generation: A Resolver then generates patches based on this structured QA knowledge.
Results
- The paper argues that this approach transforms "implicit knowledge gaps" into "explicit, reliable understanding."
- On SWE-bench Verified, it achieves up to a 4.4 percentage point improvement in Pass@1 compared to representative pre-fixing methods, with minimal additional cost and time overhead.
More from coding & agent
- A GLP1R variant may explain stronger Ozempic weight loss, and the team built an agent workflow — julia_kiseleva · 2026-07-21
- A Claude-coded Chrome extension shames you with a private jet when you open YouTube — alex_verem · 2026-07-21
- A curated TTS list for voice agents tracks latency, cancellation, and evals — mahimairaja · 2026-07-21
- Harness engineering is emerging as the execution layer for reliable AI agents — Pavan_Belagatti · 2026-07-21
- DevFest Lisbon keynote will cover Google AI Studio’s latest vibe coding and agentic AI features — gerardsans · 2026-07-21
- Daniel Hanchen’s 2-hour workshop covers open models, reward hacking and RL — danielhanchen · 2026-07-21