Multi-Perspective Agentic Program Repair via Code Property Graphs and Temporal Execution Graphs

2026-07-14 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, quick

Summary

CT-Repair is an agentic Automated Program Repair (APR) framework addressing limitations in LLM-based bug fixing, such as large execution traces and redundant patch sampling. It represents static and dynamic code evidence using queryable Code Property Graphs (CPG) and Temporal Execution Graphs (TEG), incorporating a three-stage filtering pipeline for compact TEGs. Three finite-state-machine-guided agents analyze bugs from static, dynamic, and hybrid perspectives, independently generating evidence-grounded repair strategies. A strategy-guided generation procedure instantiates candidate patches and refines strategies based on validation. Evaluated on 854 Java bugs from Defects4J v3.0, CT-Repair repaired 489 bugs in a mixed-model configuration. A GPT-5.4-mini setup fixed 388 bugs, outperforming ReinFix by 19 and RepairAgent by 30. The union of perspectives repaired 99 more bugs than any single perspective. Filtering compacted runtime evidence, reducing method scope by 94.85% and retained records by 55.97%, showing improved effectiveness without larger patch-generation budgets.

Key takeaway

For AI Engineers developing automated program repair systems, you should integrate multi-perspective agentic reasoning with structured code representations. Your systems can achieve higher repair effectiveness by leveraging Code Property Graphs and Temporal Execution Graphs, coupled with a filtering pipeline for execution traces. This approach, demonstrated by CT-Repair's 99-bug improvement from combined perspectives, allows you to enhance repair capabilities without solely relying on larger patch-generation budgets, making your solutions more efficient and robust.

Key insights

Multi-perspective agentic reasoning with structured code and execution graphs significantly improves automated program repair.

Principles

• Structured evidence enhances LLM context.
• Multi-perspective analysis improves repair strategies.
• Filtering runtime data boosts efficiency.

Method

CT-Repair constructs compact TEGs via a three-stage filtering pipeline, then uses three FSM-guided agents for static, dynamic, and hybrid analysis to generate and refine repair strategies.

In practice

• Represent code with CPGs and TEGs.
• Deploy multi-agent systems for bug analysis.
• Filter execution traces for LLM context.

Topics

• Automated Program Repair
• Large Language Models
• Code Property Graphs
• Temporal Execution Graphs
• Agentic AI
• Software Engineering

Best for: AI Scientist, Research Scientist, AI Engineer

Related on AIssential

• Evaluating Agentic Configuration Repair for Computer Networks — Agentic LLM architectures significantly improve network configuration repair efficacy and safety through iterative validation.
• Code Reasoning for Software Engineering Tasks: A Survey and A Call to Action — LLM code reasoning improves through structured CoT, execution feedback, and inference scaling, culminating in agentic systems.
• Do These Violent Delights Have Violent Ends? Measuring the Post-Merge Fate of Agentic Code — Agentic code, despite similar overall maintenance, incurs significantly higher corrective burden and introduces more vulnerabilities post-merge.
• How I Integrated DeepSeek Into NanoAgent and Why I Built a Repair Layer for Real Coding Agents — Practical AI coding agents require robust runtimes that repair imperfect model output to prevent workflow failures.
• Agentic Large Language Models for Automated Structural Analysis of 3D Frame Systems — Agentic LLMs with structured decomposition and specialized agents can reliably automate complex 3D structural analysis tasks.
• ContextSniper: AntTrail's Token-Efficient Code Memory for Repository-Level Program Repair — Precision evidence selection significantly reduces token consumption for LLM-based repository-level program repair.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.