Imagine an AI assistant that not only fixes code but also explains its reasoning, like a seasoned code reviewer. That's the promise of SpecRover, a new tool that leverages Large Language Models (LLMs) to interpret code and fix bugs with remarkable accuracy. Unlike traditional code repair tools, SpecRover doesn't just rely on tests. It delves into the "intent" behind the code, understanding the developer's original goals. SpecRover starts by examining the code structure and the bug report, much like a human developer would. It then generates a summary of the intended behavior of the problematic code sections. This is where the real magic happens. SpecRover uses these summaries, along with generated tests, to guide its patching process. It then acts as its own reviewer, scrutinizing the patch and providing feedback. This iterative process of patching and reviewing ensures higher quality and more reliable fixes. SpecRover is built upon AutoCodeRover, an open-source LLM agent. However, SpecRover significantly improves upon it by adding this intent-extraction layer. In tests on a large dataset of real-world GitHub issues (SWE-Bench), SpecRover demonstrated a substantial 50% improvement over AutoCodeRover, and resolved 31% of all tested issues. Furthermore, it does so at a modest cost – about $0.65 per issue. SpecRover isn't just about fixing bugs; it's about understanding them. By providing explanations for its patches, it offers valuable insights to developers, aiding in code maintenance and fostering trust in AI-generated solutions. This ability to combine code repair with insightful explanations marks a significant step forward in the evolution of AI-powered software development. While future work focuses on improving precision and recall, SpecRover's ability to infer code intent opens exciting possibilities for the future of automated code improvement.