Reproduce “Bug Report Specification Refinement with Trajectory Guidance for Automated Program Repair” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

S M Farah Al Fahim, Md Nakhla Rafi, Md Ahasanuzzaman, Zeyang Ma, Dong Jae Kim, Shaowei Wang, Tse-Hsun, Chen

TrajSpec refines bug reports for Automated Program Repair (APR) by leveraging the exploratory trajectory of an unverified repair agent. This system explicitly details failure mechanisms, behavioral requirements, and implementation scope, improving repair performance by 45.53% for Mini-SWE-Agent V2 with GPT-5-mini.

Key insights

  • The execution trajectory of an unverified, exploratory agent contains valuable diagnostic information that can be systematically extracted, structured, and reviewed to refine bug report specifications.
  • A hierarchical representation of evidence, linking high-level diagnostic conclusions to mid-level reasoning and low-level code observations, is crucial for managing the noise and length of raw trajectories.
  • Repository-based review is critical for validating claims derived from unverified agent runs, preventing the propagation of incorrect or weakly supported information into the refined task specification.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.