AI agent 写的 React，它来体检：确定性扫描 + Agent 技能 + CI 门禁

React Doctor is a deterministic React code scanner that you run at your project root to get an audit. It is built for developers and AI coding agents: once the audit is available, the same tool can install an agent skill so your assistant learns from the issues and fixes them going forward. The project is MIT-licensed and maintained under the millionco umbrella.

LLM-based reviews can be inconsistent; React Doctor deliberately scans deterministically. It catches issues across state and effects, performance, architecture, security, accessibility, and maintainability. It also points out overly complex React functions and repeated JSX trees as composition candidates, giving agents and humans concrete targets to refactor.

The scanner is rule-based, so the same code always produces the same findings. Rules span multiple concerns rather than just one linter dimension. The repo structure shows it ships both an ESLint plugin (packages/eslint-plugin-react-doctor) and an oxlint plugin (packages/oxlint-plugin-react-doctor), with a shared core (packages/core) and a public CLI (packages/react-doctor).

Once you have an audit, run npx react-doctor@latest install to register a skill for your coding agent. The README lists Claude Code, Cursor, Codex, OpenCode, and more. This turns a one-off audit into an ongoing capability: your agent can consult the findings and avoid repeating them.

React Doctor reviews every pull request and reports only the issues your change introduced, not your existing backlog. npx react-doctor@latest ci install adds the workflow, scans every pull request, and posts a summary comment. You can tune gate, scope, and comments with react-doctor ci config, and bump the action with react-doctor ci upgrade. GitHub Actions is fully supported; GitLab CI gets a gate-only scaffold.

Beyond static analysis, React Doctor can capture a Chrome DevTools performance trace while you interact with a running app. npx react-doctor@latest scan http://localhost:3000 opens system Chrome in an isolated profile, records until you press Enter (max five minutes), and flashes purple outlines with component names as React renders. The output is a readable summary plus the path to a compressed trace. For authenticated sessions, start a dedicated Chrome profile with remote debugging and pass its endpoint via --cdp. The trace stays local and is never uploaded, but it may contain page URLs, source paths, and React profiling details, so treat it as sensitive.

The quickest start is a one-liner: npx react-doctor@latest at your project root. From there, the flow branches by need: install the agent skill (npx react-doctor@latest install), set up CI (npx react-doctor@latest ci install), or scan a running app. Rules can be configured in doctor.config.ts.

React Doctor fits teams that want consistent, deterministic React checks integrated into agents and CI, across Next.js, Vite, Astro, TanStack, React Native, Expo, and more. Keep in mind: telemetry is on by default (CLI version, commands, framework, React version, project size, rule counts — no file contents) and can be disabled with npx react-doctor@latest --no-telemetry. In an interactive terminal you can run react-doctor scan without a URL and pick a local app; agents and CI must pass the URL explicitly. Because it is rule-based, it catches concrete anti-patterns rather than offering free-form LLM judgment.