HGARena

This repository is a public example release of HGARena centered on the astropy__astropy-12907 case.

HGARena is a repository-level issue resolution framework that combines:

• heterogeneous repository graph construction,
• lexical-first retrieval with graph reranking,
• iterative submitter--reviewer--executor search, and
• Docker-based SWE-bench evaluation.

Repository Layout

• framework/

  • hgarena/: core runtime code for graph construction, retrieval, arena search, and task execution.
  • prompts/: prompt templates for the planner, implementer, and reviewer.
  • examples/configs/: example configuration for running the public case.
  • data/: SWE-bench Lite-oriented example input.
  • main.py: entry point for running the example pipeline.

• records/12907_success/

  • One successful GPT-4o record for astropy__astropy-12907.

• records/12907_failure/

  • One failed GPT-4o record for astropy__astropy-12907.

Example Scope

This release is intentionally limited to the astropy__astropy-12907 example.

Both record directories correspond to GPT-4o runs on the same task:

• records/12907_success/: one success case
• records/12907_failure/: one failure case