code-split-pr

What problem does it solve? Oversized pull requests are hard to review, risky to merge, and painful to revert. This Skill turns one large donor diff into an ordered plan of small PRs, each encoding a single human decision or an isolated mechanical change, so reviewers can judge one question at a time. ## Core Features & Use Cases - Decision vs. mechanical classification: Separates PRs that ask reviewers one product or architecture question from behavior-preserving code motion such as moves, renames, and dead-code deletion. - Coverage map and ordering: Assigns every changed donor file to exactly one batch and orders batches so each intermediate stack state stays green and shippable. - Rebuild strategy and executor handoff: Recommends fresh-stack-from-trunk versus in-place splitting, with diff-to-zero verification against the donor. - Use Case: You have a 3,000-line feature branch mixing refactors, renames, and one real behavior change. The Skill mines the diff for learnings, proposes a stack of reviewable PRs with reviewer questions, and iterates the plan with you before any branch is touched. ## Quick Start Ask the agent to plan how to split the current oversized branch or PR into a stack of small decision-sized PRs without modifying any branches.