What problem does it solve? Large pull requests that mix hand-written contracts, generated code, behavior fixes, and tests force reviewers to dig through hundreds of irrelevant lines to find the changes that matter. This Skill splits one change into a stack of PRs by meaning, so each reviewer sees only the layer that needs their scrutiny. ## Core Features & Use Cases - Semantic Layer Classification: Assigns every changed file to exactly one of four layers — Base (codegen input), Generated (generator output), Fix (behavior change), and Tests — using evidence like generated-file headers and lockfile patterns. - Automated Stack Building: Creates branches, commits, and draft PRs bottom-up with gh stack, then rewrites each PR with a proper title, body, and label (generated, test). - Drift-Check Awareness: Detects CI codegen drift checks that would make a Base-only PR fail, and offers to fold Base and Generated into one PR with two commits. - Use Case: You changed an OpenAPI spec, regenerated the client (800 lines), fixed retry logic (20 lines), and added tests. The Skill produces four stacked PRs so reviewers read the contract and the fix while skipping the generated diff entirely. ## Quick Start Split my current uncommitted changes into a stacked PR by meaning using the semantic-stacked-pr skill.
