What problem does it solve? Coordinating large-scale refactoring across a pull request is risky: a single reviewer or model may miss structural issues, behavior changes can slip in unnoticed, and there is no systematic way to converge on a stable set of improvements. This Skill distributes proposal, application, and verification work across multiple AI CLI runtimes (claude, codex, agy, kiro) and loops until no new refactoring proposals appear. ## Core Features & Use Cases - Multi-runtime convergence loop: Three CLIs propose structural improvements, proposals are deduplicated and prioritized, then applied in non-overlapping groups with one commit per apply round. - Test-first safety net: A test-provisioning round adds characterization tests before any structural change, and every apply round is verified by running the baseline test command; failing groups are reverted. - Mechanical verification: Commit trailers, diff budgets, scope enforcement, and test expectation checks are validated from git and real test runs rather than agent self-reports. - Use Case: Given draft PR #130 touching src/services, run the skill with a scope and baseline test command so that codex, agy, and kiro propose refactorings, apply them round by round, and finish with a cross-review gate and a metrics report. ## Quick Start Ask the AI to run /ndf:cross-refactoring on your draft pull request with a scope like src/services plus tests/services and a baseline test command such as pytest -q.
