What problem does it solve?
Split a large GitHub pull request into smaller PRs so that each new PR requires fewer CODEOWNERS reviewer groups, reducing review burden and speeding up approval.
Core Features & Use Cases
- CODEOWNERS-aware file grouping: Analyze changed files and map them to CODEOWNERS owner groups to understand which reviewers each part of the PR will involve.
- Optimization to minimize reviewer groups: Propose a split that clusters files by their required reviewer groups, using the largest cluster first and grouping the rest to keep reviewer sets small.
- Dependency-aware sequencing: Explicitly handle cases where one split PR must be merged before another due to renames or symbol dependencies, ensuring each PR can be merged cleanly to main.
Quick Start
Use the split-pr skill to split PR number you specify by analyzing NVIDIA/Megatron-LM CODEOWNERS, proposing an optimized multi-PR plan, and then creating draft PRs on your fork after you approve the proposed split.