propose-refactor

A codebase drift analysis scans for duplicated logic, mixed concerns, and unreferenced files, then writes up to three prioritized refactor proposals to .agent/queue.md without modifying source files. It preserves safety by never executing changes during the audit.

The best way to prioritize refactor tasks is by scoring findings based on Impact and Effort, surfacing up to three proposals per run. This Skill ranks drift signals and code smells to help focus on high-impact, manageable refactors first during major version prep.

To generate a markdown queue of refactor proposals, run a repository-scoped scan that formats findings into prescribed markdown blocks with metadata. The Skill writes these proposals, complete with rollback and validation guidance, directly to .agent/queue.md.

Yes, this codebase audit tool supports nightly automated drift detection triggered by scheduled audits or scaffolds. It performs per-run drift detection, analyzes the codebase, and outputs formatted refactor proposals to the queue file without executing changes.

A key limitation of automated refactor proposal generation is that it is strictly limited to three proposals per run and never executes refactors or modifies code. It only surfaces actionable proposals with rollback guidance, requiring manual implementation afterward.