When your coding agent writes bad React, send it to the doctor

What it does

React Doctor is a deterministic scanner that audits React codebases for sins in state, effects, performance, architecture, security, and accessibility. It runs locally and surfaces findings for human review in the terminal or as inline CI annotations. The hook is that it also packages those findings into an agent “skill,” so Claude Code, Cursor, Codex, and others can—in theory—learn from the audit and stop repeating the same anti-patterns.

The interesting bit

Most static analysis tools scold the developer; this one scolds the developer and their robot intern. By turning audit results into a reusable skill, it tries to close the feedback loop between static analysis and generative coding. It is a neat admission that the fastest-growing consumer of React code is no longer human.

Key highlights

• Covers Next.js, Vite, TanStack, React Native, Expo, and plain React.
• GitHub Action posts inline annotations on pull requests where reviewers already look.
• Exports skills for Claude Code, Cursor, Codex, and OpenCode.
• Rules are configurable in doctor.config.ts (or JSON/JS variants).
• Telemetry is limited to metadata and rule counts by default; the README explicitly states no file contents or specific findings are transmitted.

Caveats

• Telemetry is on by default and the opt-out is a per-run CLI flag; the README does not describe a persistent way to disable it.
• The README does not quantify the rule set or expected false-positive rate, so it is unclear how noisy the scanner is on large codebases.

Verdict

Grab it if you are running AI agents against a React codebase and want automated quality control with a feedback loop. Pass if you need a general-purpose linter—this is tightly scoped to React and its frameworks.