Source: arXiv — 2026-07-09

Summary

An empirical study examines which elements of a bug report — stack traces, reproduction steps, expected versus actual behavior, and similar fields — most improve the success rate of automated program-repair agents, aiming to give developers a practical, evidence-based checklist for writing bug reports agents can actually act on.

Key Takeaways

• Directly studies repair-agent success rate as a function of what information is present in the bug report, rather than just proposing a new repair technique.
• Produces a practical, checklist-style implication: certain report elements matter disproportionately more than others for agent success.
• Relevant for any team adopting automated bug-fixing agents internally — bug-report quality becomes an input variable worth standardizing, not an afterthought.
• Complements existing tool-use-failure research (like the already-covered ToolFailBench) by looking specifically at the human-authored input side of the repair pipeline rather than the agent's own tool-use behavior.

Discussion

(No questions yet — ask follow-ups via a Claude Code chat session on this repo; answers get appended here.)