shadcn/improve

Source repo for the improve agent skill: audit a codebase, vet findings, and write self-contained implementation plans for other agents to execute.

Routing Summary

Use shadcn/improve when the expensive work is diagnosis, prioritization, or plan authorship. It should create auditable improvement plans that cheaper agents can execute later, then hand completed work to Security and Review Skills and Security and Review Skills for hardening.

What it is

shadcn/improve is the upstream source for Kevin's installed Frontend and Design Skills skill. Its core move is Borrowed Intelligence (Drain It into Plans): use the strongest model for understanding, judgment, prioritization, and plan writing; hand execution to cheaper agents later. The skill writes to plans/ and deliberately does not implement source changes itself.

The follow-up bookmark adds the model-risk rationale. Shadcn framed /improve as a hedge after a frontier model was pulled: the best model today may disappear, be priced differently, or be replaced tomorrow. The durable move is to convert temporary model advantage into repo-specific plans, issues, and backlog artifacts while the model is available. Source: X/@shadcn, 2026-06-13

Install shape:

npx skills add shadcn/improve

Routing

Need Route
Audit a repo and bank high-leverage plans /improve / Frontend and Design Skills
Scope to current branch /improve branch
Focus on security, perf, tests, bugs, etc. /improve security, /improve perf, etc.
Execute a banked plan Browser Testing Skills or /improve execute <plan> with review
Strict final PR review Security and Review Skills then Security and Review Skills

The boundary matters: improve is for plan creation and plan governance. It is not a replacement for the implementation loop or review doctors.

Version Snapshot

Field Snapshot
Repo shadcn/improve
Release status No GitHub releases observed as of 2026-07-04
Package status Installed through npx skills add shadcn/improve; no npm package used as version authority
License MIT
Repo status 6,823 stars / 274 forks / 10 open issues as of 2026-07-04
Last repo push checked 2026-06-15
Default-branch HEAD 03369ee6d7cafbfcecc4346539b05b3dc0a603bb

Source: GitHub API and README, 2026-07-04

Plan Quality Invariants

Every accepted plan should be:

  • self-contained enough for a weaker executor
  • stamped to a git commit for drift checking
  • explicit about files, current state, steps, verification, and STOP conditions
  • scoped so the executor does not improvise product decisions
  • reconciled later so stale plans do not accumulate

Kevin Stack Placement

In Kevin's stack, shadcn/improve lives before PRD/code for broad codebase improvement and after initial domain/context discovery. Pair it with Frontend and Design Skills for architecture questions, Agent Self-Improvement Eval Library when improving the agent loop itself, and Security and Review Skills when the resulting implementation is ready to harden.


Timeline