Problem
Research questions and friction points this paper is trying to address.
CI repair
experience reuse
problem-solution alignment
pull request
software history
RETRACE: From Entangled Repair Histories to Reusable Experience for CI Repair
🤖 AI Summary
This study addresses the misalignment between continuous integration (CI) fixes and their corresponding issues in software development histories, which hinders the reuse of past repair experiences. To overcome this challenge, we propose a bidirectional view fusion mechanism that reconstructs issue-level repair knowledge by integrating endpoint backtracking and developmental evolution perspectives, grounded in CI execution evidence. Specifically, our approach leverages large language model (LLM) agents, a bidirectional reasoning framework, and multi-granularity experience representations to achieve hierarchical abstraction from unstructured commit histories into transferable repair patterns. Evaluated on the CI-REPAIR-BENCH benchmark, the proposed method yields substantial improvements in the Pass@1 metric, achieving 31.9% with MiniMax-M2.5 and 32.8% with DeepSeek-V4-Flash, demonstrating its effectiveness in automated CI fix generation.
📝 Abstract
Large language model (LLM) agents increasingly reuse prior experience, but most approaches assume that problems and solutions are already aligned. Software histories rarely provide this alignment: a pull request (PR) may contain multiple continuous integration (CI) problems, failed attempts, reverted edits, and unrelated changes, obscuring which changes resolve each problem. We present RETRACE, a framework for reconstructing problem-level repair experience from such histories. RETRACE combines an endpoint view that reasons backward from changes retained in the passing revision with a development view that traces repair evolution forward through commit history. CI execution evidence reconciles the two views, and the recovered experience is represented at three abstraction levels, from concrete fixes to transferable repair patterns. For new failures, RETRACE retrieves relevant problem-level experience to guide repair. On CI-REPAIR-BENCH, comprising 565 PR-level repairs from 101 repositories across 12 failure categories, RETRACE improves mini-SWE-agent Pass@1 from 19.6% to 31.9% with MiniMax-M2.5 and from 23.3% to 32.8% with DeepSeek-V4-Flash. On a matched subset, Codex improves from 15.5% to 27.5%. Combining both views consistently outperforms either alone, showing that recovering problem-change alignment enables historical CI repairs to serve as reusable repair experience.
Innovation
Methods, ideas, or system contributions that make the work stand out.
CI Repair
Experience Reuse
Problem-Change Alignment
LLM Agents
Repair History Reconstruction
🔎 Similar Papers
No similar papers found.
