---
title: "CT-Repair: Multi-Perspective Agentic Program Repair via Code Property Graphs and Temporal Execution Graphs"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.12605v1"
permalink: "https://feed.astralor.com/items/itm_92bf1f427a77b7cedc2588cc"
published: "2026-07-14T10:33:29.000Z"
score: 81
channel: "ai"
tags: ["CT-Repair","GPT-5.4-mini","Code Property Graph","Multi-Agent","Temporal Execution Graph","Automated Program Repair"]
---

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

## AI commentary

> What impresses me most about CT-Repair is its philosophy: 'Don't just increase the number of patches, let the model see the soul of the program'—by using CPG and TEG to let LLMs understand the static skeleton and dynamic lineage of code. For researchers and engineers in AI for Software Engineering, this paper offers not only a reproducible high-performance baseline but also a clever integration of program analysis tools with LLMs, especially valuable in context-window-sensitive scenarios. I highly recommend reproducing the experiments with their open-source toolchain.

## At a glance

Large language models (LLMs) have improved automated program repair (APR), but face limitations such as repetitive execution traces and lack of root-cause diversity in patch sampling. This paper presents CT-Repair, an agentic APR framework that represents static and dynamic evidence as queryable Code Property Graphs (CPG) and Temporal Execution Graphs (TEG). A three-stage filtering pipeline constructs compact TEGs, and three finite-state-machine-guided agents analyze each bug from static, dynamic, and hybrid perspectives to produce evidence-grounded repair strategies. Strategy-guided generation instantiates these strategies as candidate patches and refines the most promising one using validation feedback. Evaluated on 854 Java bugs from Defects4J v3.0, CT-Repair correctly repairs 489 bugs in a mixed-model configuration, and under a controlled GPT-5.4-mini configuration, it repairs 388 bugs, surpassing ReinFix and RepairAgent by 19 and 30 respectively. The union of the three perspectives repairs 99 more bugs than the strongest individual perspective. The filtering pipeline compacts runtime evidence, with execution filtering narrowing candidate method scope by 94.85% on average and behavior filtering further reducing retained runtime records by 55.97%. These results demonstrate that structured runtime evidence and multi-perspective reasoning can improve repair effectiveness without solely relying on a larger patch-generation budget.

## Key points

- CT-Repair unifies static and dynamic program evidence via Code Property Graphs (CPG) and Temporal Execution Graphs (TEG), enabling multi-perspective reasoning.
- A three-stage filtering pipeline (execution filtering, behavior filtering, TEG construction) compresses runtime evidence to under 5% of its original size, drastically improving context efficiency.
- Three finite-state-machine-guided agents generate repair strategies from static, dynamic, and hybrid perspectives; strategy-directed patch generation is refined via validation feedback.
- On Defects4J v3.0, CT-Repair with GPT-5.4-mini correctly repairs 388 bugs, surpassing ReinFix and RepairAgent, and the union of perspectives yields 99 more fixes than the strongest single perspective.

## Full text

Large language models (LLMs) have improved automated program repair (APR), but two limitations remain. First, raw execution traces are often too large and repetitive to serve as effective model context. Second, repeated patch sampling may produce different implementations without yielding distinct root-cause hypotheses or repair strategies. We present CT-Repair, an agentic APR framework representing static and dynamic evidence as queryable Code Property Graph (CPG) and Temporal Execution Graph (TEG). CT-Repair applies a three-stage filtering pipeline to construct compact TEGs. Three finite-state-machine-guided agents analyze each bug from static, dynamic, and hybrid perspectives and independently produce evidence-grounded repair strategies. A strategy-guided generation procedure instantiates these strategies as candidate patches and uses validation feedback to refine the most promising strategy. We evaluate CT-Repair on 854 Java bugs from Defects4J v3.0. In the mixed-model configuration, CT-Repair correctly repairs 489 bugs. Under a controlled GPT-5.4-mini configuration, it repairs 388 bugs, 19 and 30 more than ReinFix and RepairAgent, respectively. The union of the three evidence perspectives repairs 99 more bugs than the strongest individual perspective. The filtering pipeline also compacts runtime evidence, with execution filtering narrowing the candidate method scope by 94.85% on average and behavior filtering further reducing retained runtime records by 55.97%. These results show that structured runtime evidence and multi-perspective reasoning can improve repair effectiveness without relying solely on a larger patch-generation budget.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_92bf1f427a77b7cedc2588cc