Replication Package: Can Coding Agents Produce Merge-Ready Feature Pull Requests? An Industrial Evaluation of 62 Real Tasks

Replication package for the ASE 2026 Industry Showcase paper. Contains 62 feature-implementation task definitions (prompts, ground-truth diffs, file classifications), a task registry with commit SHAs and PR URLs, the scoring rubric, evaluation scores for 496 agent runs across four commercial coding agents (Claude Code, Codex CLI, Cursor, OpenCode), and evaluation/analysis scripts.适用于ASE 2026行业展示论文的复现包，包含62项功能实现任务的定义（含提示词、基准真值代码差异（ground-truth diffs）、文件分类）、带有提交哈希值（commit SHAs）与拉取请求链接（PR URLs）的任务注册表、评分细则，以及覆盖Claude Code、Codex CLI、Cursor、OpenCode四款商业化代码智能体的496次智能体运行评估得分，同时附带评估与分析脚本。