Autonomous Task Completion on GitHub
84Success RateGithub + GitLab (PolySkill)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Github + GitLab (PolySkill)Training Setting=Self-guided Exploration, Iters=1002025.10 | 84 | 39.5 | |
| GithubTraining Setting=Single-Domain Specialists, Iters=502025.10 | 81.5 | 54.1 | |
| Gitlab → GithubTraining Setting=Sequential Curriculum, Iters=50 + 502025.10 | 80.1 | 51.9 | |
| Github → GitLabTraining Setting=Sequential Curriculum, Iters=50 + 502025.10 | 77.8 | 48.6 | |
| GitLabTraining Setting=Single-Domain Specialists, Iters=502025.10 | 71.5 | 12.8 | |
| BaselineIters=–2025.10 | 66.7 | — |