---
dataset_info:
  features:
  - name: task_id
    dtype: string
  - name: repo
    dtype: string
  - name: pr_number
    dtype: int32
  - name: base_sha
    dtype: string
  - name: head_sha
    dtype: string
  - name: task
    dtype: string
  - name: plan
    list:
    - name: subtask_id
      dtype: string
    - name: goal
      dtype: string
    - name: scope_files
      list: string
  - name: subagents
    list:
    - name: subtask_id
      dtype: string
    - name: scope_files
      list: string
    - name: solution_diff
      dtype: string
    - name: summary
      dtype: string
  - name: revisions
    list:
    - name: subtask_id
      dtype: string
    - name: comment
      dtype: string
    - name: revised_diff
      dtype: string
  - name: final_diff
    dtype: string
  - name: num_subtasks
    dtype: int32
  - name: num_changed_files
    dtype: int32
  splits:
  - name: train
    num_bytes: 7867623532
    num_examples: 34696
  download_size: 2326027093
  dataset_size: 7867623532
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
tags:
- code
- agents
- github
- swe
- trajectories
- planning
license: other
language:
- code
- en
size_categories:
- 10K<n<100K
task_categories:
- text-generation
pretty_name: GitHub PR Agent Trajectories
---


## Description

Reconstructed multi-step agent trajectories for resolving public GitHub PRs: a `plan` of subtasks and per-subtask `subagents` with scoped files and solution diffs.

> Derived dataset. Source material retains its original per-item licence (see `source`/`repo` columns); treat as **other / mixed**. Provided as-is.

## Usage

```python
from datasets import load_dataset
ds = load_dataset("PotatoHD/github-pr-agent-trajectories")
```