Synthetic GitHub Issues and Agent Trajectories for Python

SWE-Lego-Synthetic-Data contains 11.5k synthetic GitHub issues written in Python and their multi-turn agent trajectories. The data was generated by the SWE-Lego project using the Qwen/Qwen3-Coder-480B-A35B-Instruct model with OpenHands agent scaffolding. It was last updated on January 8, 2026.

Use Cases

• Supervised fine-tuning (SFT) of language models using the `messages` column containing agent trajectories.
• Training or benchmarking code-generation agents on multi-turn interactions derived from synthetic GitHub issues.
• Analyzing patterns in simulated software engineering workflows captured in the issue and trajectory data.
• Studying the behavior of AI agents scaffolded with OpenHands (v0.53.0) on Python programming tasks.

Strengths

• Contains 11,500 synthetic data points, providing a substantial corpus for training.
• Includes multi-turn agent trajectories, offering sequential interaction data for complex tasks.
• Data is directly formatted for SFT training, reducing initial preprocessing effort.

Limitations

• Data is synthetic, generated by a model, and may not fully reflect real-world GitHub issue patterns.
• Limited to Python language issues, restricting applicability to other programming ecosystems.
• The dataset's scope is confined to the specific agent scaffolding (OpenHands v0.53.0) used for generation.

Provenance

Source

SWE-Lego project, hosted on Hugging Face.

Collection Method

Generated synthetically using the Qwen/Qwen3-Coder-480B-A35B-Instruct model with OpenHands (v0.53.0) agent scaffolding.

Freshness

Last updated on January 8, 2026.