Hi there 👋

RepoU is a ✨ Official implementation ✨ repository of paper "RepoUnderstander: Enhancing Automated Issue Resolving with Comprehensive Repository Knowledge Exploration".

• Recently, Large Language Model (LLM) based agents have advanced the significant development of Automatic Software Engineering (ASE).
• We develop a novel ASE method named RepoUnderstander by guiding agents to comprehensively understand the whole repositories.
• It achieved 18.5% relative improvement on the SWE-bench Lite benchmark compared to SWE-agent.

1. Before running, you should make sure you have the SWE-bench environment installed.

See https://github.com/princeton-nlp/SWE-bench

2. Configure OpenAI token

RepoUnderstaner/app/model/gpt.py

3. Run our code:

python scripts/run.py conf/vanilla-lite.conf

4. MCTS Code

RepoUnderstaner/app/MCTS/

5. FIX DATASET

fix_dataset.jsonl

format (jsonl):
- instance_id
- signiture_changed_by_rule
- has_additional_info
- additional_info
- problem_statement
- patch

Acknowledgements

Thanks for the following open-source projects:

• SWE-bench
• SWE-agent
• AutoCodeRover