shadcn's MIT-licensed agent skill that audits a codebase with your most capable model and writes self-contained implementation plans for cheaper models to execute. It never implements anything itself; the plan is the product.
Run /improve in a repo and it maps the stack, conventions and the exact build, test and lint commands, then fans out subagents across nine categories: correctness, security, performance, test coverage, tech debt, dependencies, DX, docs and direction. Every finding carries file:line evidence, impact, effort and confidence. The advisor re-reads each cited location before showing you the table, drops false positives and records rejections so they don't come back on the next run.
You pick the findings you want, and each becomes a plan in plans/ with the current code excerpted, exact steps, the repo's own commands as verification gates and STOP conditions. Hand a plan to any agent, or run /improve execute <plan> to dispatch a cheaper model in an isolated worktree and review its diff against the plan. Other modes include quick, deep, branch (only what the current branch changes), next (feature suggestions) and reconcile (refresh the backlog).
Install with npx skills add shadcn/improve. It works in any agent that supports the Agent Skills format.