[GitHub Trending] shadcn/improve

8.7 relevance

Shadcn's tool for using powerful models to audit code and generate plans for cheaper models, highly innovative and actionable for AI agent workflows.

AI/ML github.com

Answer the board on AI readiness

Arrive with evidence, not opinion. See how your foundation benchmarks against your peers, get a clear read, and leave with a ranked place to start.

Why It Matters

For platform engineers and AI agent orchestrators, this introduces a practical separation of concerns pattern — using expensive models for architecture-level reasoning and cheap models for implementation — that directly addresses the cost-quality tradeoff in multi-agent systems.

Key Takeaways

Adopt the /improve pattern in your agent workflows: route code understanding and planning to your most capable model, then dispatch execution to cheaper models using self-contained markdown specs as the contract.

Use your most capable model to audit your codebase and write plans for cheaper models to execute. - shadcn/improve

Summary

shadcn/improve is an agent skill that audits any codebase and produces self-contained markdown implementation plans for other agents to execute, separating expensive reasoning from cheap execution. It maps the repo's stack, conventions, and build commands, then fans out parallel subagents across nine categories (correctness, security, performance, test coverage, tech debt, dependencies, DX, docs, direction) to produce prioritized findings with confidence scores. Plans include code excerpts, exact steps, verification gates using the repo's own test/lint commands, and STOP conditions, with commands for scoping audits to branches, reconciling backlogs, and dispatching cheaper executors in isolated worktrees.