A Survey of LLM-based Automated Program Repair: Taxonomies, Design Paradigms, and Applications

Abstract: LLMs are reshaping automated program repair. We present a unified taxonomy that groups 62 recent LLM-based repair systems into four paradigms defined by parameter adaptation and control authority over the repair loop, and overlays two cross-cutting layers for retrieval and analysis augmentation. Prior surveys have either focused on classical software repair techniques, on LLMs in software engineering more broadly, or on subsets of LLM-based software repair, such as fine-tuning strategies or vulnerability repair. We complement these works by treating fine-tuning, prompting, procedural pipelines, and agentic frameworks as first-class paradigms and systematically mapping representative systems to each of these paradigms. We also consolidate evaluation practice on common benchmarks by recording benchmark scope, pass@k, and fault-localization assumptions to support a more meaningful comparison of reported success rates. We clarify trade-offs among paradigms in task alignment, deployment cost, controllability, and ability to repair multi-hunk or cross-file bugs. We discuss challenges in current LLM-based software repair and outline research directions. Our artifacts, including the representation papers and scripted survey pipeline, are publicly available at https://github.com/GLEAM-Lab/ProgramRepair.

• The paper introduces a unified taxonomy of LLM-based program repair paradigms by categorizing systems based on parameter adaptation and control authority.
• It evaluates 62 repair systems using benchmarks and pass@k metrics to highlight trade-offs in deployment cost, controllability, and bug resolution.
• The study identifies challenges in evaluation protocols, prompt sensitivity, and scalability, and outlines future directions for more robust repair solutions.

A Survey of LLM-based Automated Program Repair: Taxonomies, Design Paradigms, and Applications

Introduction

The paper "A Survey of LLM-based Automated Program Repair: Taxonomies, Design Paradigms, and Applications" (2506.23749) comprehensively examines the integration of LLMs within the domain of automated program repair. The work presents a unified taxonomy that categorizes LLM-based repair systems into four distinct paradigms based on parameter adaptation and control authority, and overlays two cross-cutting layers for retrieval and analysis augmentation. This taxonomy provides a structured approach to understanding the diverse methodologies employed in LLM-based repair.

Taxonomy and Design Framework

The authors introduce a design framework that organizes LLM-based program repair into four paradigms: fine-tuning, prompting, procedural pipelines, and agentic frameworks. Each paradigm is defined by the degree to which parameters are adapted and the control authority maintained over the repair loop. The taxonomy further distinguishes these paradigms by the nature of context augmentation through Retrieval-Augmented Generation (RAG) and Analysis-Augmented Generation (AAG) techniques.

Figure 1: Taxonomy of LLM-based Software Repair: four paradigms defined by parameter updates and control authority, with retrieval- and analysis-augmented contexts as cross-cutting layers.

This new taxonomy allows for an explicit comparison of trade-offs among the paradigms in terms of task alignment, deployment cost, controllability, and the ability to address multi-hunk or cross-file bugs. The framework provides insights into the strengths and weaknesses of current LLM-based software repair systems and how they are deployed across different scenarios, ranging from educational settings to industrial applications.

Survey of Current Systems

In this survey, 62 LLM-based repair systems are evaluated and categorized. These systems are analyzed based on their benchmark scope, pass@k success metrics, and fault-localization assumptions, providing a solid foundation for comparing reported success rates across different repair strategies. The paper meticulously records and contrasts the evaluation practices employed by various systems, clarifying often-overlooked differences in benchmarks and metrics used.

Figure 2: Overview of LLM-based software repair techniques: parallel columns show workflow control within each paradigm, and the left bar overlays retrieval and analysis augmentation.

The study maps these systems into the proposed taxonomy and evaluates their performance on common benchmarks, highlighting the distinct operational settings and constraints involved in each approach. By addressing the nuances of these systems, the paper provides clear guidance on the appropriateness of different paradigms for varied software repair tasks.

Challenges and Future Directions

The research identifies several ongoing challenges, including the training and deployment costs associated with LLMs, the reliability of evaluation protocols, and the robustness of workflows against prompt sensitivity and variability. The survey emphasizes the need for stronger benchmarks and more transparent reporting of evaluation metrics to better assess the effectiveness of different paradigms.

Figure 3: Workflow of literature survey.

The authors speculate on future research opportunities in enhancing the precision of context retrieval and selection, improving evaluation frameworks, and integrating more robust mechanisms for semantic verification of patches. By addressing these challenges, the field can advance towards more practical and scalable LLM-based repair systems.

Conclusion

This survey categorizes LLM-based program repair techniques using a novel taxonomy and highlights significant advancements in the field. It provides a comprehensive examination of current methodologies, evaluates their efficacy across various benchmarks, and identifies key areas for future research. The paper serves as a vital resource for researchers and practitioners seeking to understand the landscape of LLM-based software repair and its practical implications.