Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SIADAFIX: issue description response for adaptive program repair

Xin Cao, Nan Yu

TL;DR

This work tackles the challenge of unstable, complex repository-level program repair by introducing SIADAFIX, an adaptive framework that marries fast, intuition-driven issue-description processing with slow, multi-step bug-fix reasoning. It employs a two-phase architecture: (i) a Workflow Decision Phase with an Issue Description Optimizer and a RandomForest-based Issue Description difficulty Classifier to tailor repair strategies, and (ii) a Workflow Execution Phase comprising a Bug Fix Agent, a Multi-level Checker System, and a Selector to realize Easy, Middle, and Hard repair modes. The approach is evaluated on SWE-bench Lite using Claude-4 Sonnet, achieving a strong $60.7\%$ Pass@1 and demonstrating state-of-the-art performance among open-source methods. The results highlight the value of adaptive orchestration for efficient and accurate automated program repair, with future work aimed at architecture analysis, multi-modal data fusion, and domain-aware knowledge integration.

Abstract

We propose utilizing fast and slow thinking to enhance the capabilities of large language model-based agents on complex tasks such as program repair. In particular, we design an adaptive program repair method based on issue description response, called SIADAFIX. The proposed method utilizes slow thinking bug fix agent to complete complex program repair tasks, and employs fast thinking workflow decision components to optimize and classify issue descriptions, using issue description response results to guide the orchestration of bug fix agent workflows. SIADAFIX adaptively selects three repair modes, i.e., easy, middle and hard mode, based on problem complexity. It employs fast generalization for simple problems and test-time scaling techniques for complex problems. Experimental results on the SWE-bench Lite show that the proposed method achieves 60.67% pass@1 performance using the Claude-4 Sonnet model, reaching state-of-the-art levels among all open-source methods. SIADAFIX effectively balances repair efficiency and accuracy, providing new insights for automated program repair. Our code is available at https://github.com/liauto-siada/siada-cli.

SIADAFIX: issue description response for adaptive program repair

TL;DR

This work tackles the challenge of unstable, complex repository-level program repair by introducing SIADAFIX, an adaptive framework that marries fast, intuition-driven issue-description processing with slow, multi-step bug-fix reasoning. It employs a two-phase architecture: (i) a Workflow Decision Phase with an Issue Description Optimizer and a RandomForest-based Issue Description difficulty Classifier to tailor repair strategies, and (ii) a Workflow Execution Phase comprising a Bug Fix Agent, a Multi-level Checker System, and a Selector to realize Easy, Middle, and Hard repair modes. The approach is evaluated on SWE-bench Lite using Claude-4 Sonnet, achieving a strong Pass@1 and demonstrating state-of-the-art performance among open-source methods. The results highlight the value of adaptive orchestration for efficient and accurate automated program repair, with future work aimed at architecture analysis, multi-modal data fusion, and domain-aware knowledge integration.

Abstract

We propose utilizing fast and slow thinking to enhance the capabilities of large language model-based agents on complex tasks such as program repair. In particular, we design an adaptive program repair method based on issue description response, called SIADAFIX. The proposed method utilizes slow thinking bug fix agent to complete complex program repair tasks, and employs fast thinking workflow decision components to optimize and classify issue descriptions, using issue description response results to guide the orchestration of bug fix agent workflows. SIADAFIX adaptively selects three repair modes, i.e., easy, middle and hard mode, based on problem complexity. It employs fast generalization for simple problems and test-time scaling techniques for complex problems. Experimental results on the SWE-bench Lite show that the proposed method achieves 60.67% pass@1 performance using the Claude-4 Sonnet model, reaching state-of-the-art levels among all open-source methods. SIADAFIX effectively balances repair efficiency and accuracy, providing new insights for automated program repair. Our code is available at https://github.com/liauto-siada/siada-cli.
Paper Structure (56 sections, 3 figures, 4 tables)

This paper contains 56 sections, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Workflow Decision Phase
  5. Issue Description Optimizer
  6. Issue Description difficulty Classifier
  7. Workflow Execution Phase
  8. Bug Fix Agent
  9. Multi-level Checker System
  10. Selector
  11. Three Repair Modes
  12. Experiments
  13. Experimental Setup
  14. Dataset
  15. Baselines
  16. ...and 41 more sections

Figures (3)

  • Figure 1: Ablation study of SIADAFIX. BFA: slow thinking Bug fix agent; BFA+CH: BFA enhanced by fast thinking checker; BFA+OP: BFA with fast thinking issue description Optimizer; BFA+CH+OP: BFA combined with issue description Optimizer and checker in single workflow mode; BFA+CH+OP+CL: The proposed SIADAFIX including adaptive Classifier (CL) for three modes, i.e. Easy, Middle and Hard.
  • Figure 2: Overview of the SIADAFIX framework. The framework consists of two main phases: (1) Workflow Decision Phase with Issue Description Optimizer and Classifier, and (2) Workflow Execution Phase with bug fix agent, Checkers, and Selectors operating in Easy, Middle, and Hard modes.
  • Figure 3: Overview of the Bug FIX Agent.