Compare the changed files with the original task. Watch for unrelated refactors, formatting churn, dependency edits, or generated files that were not part of the request.
Which files changed outside the original task, and why were they necessary?
AI 工作流工具
AI 代码审查清单
在合并或部署前，审查 Cursor、Claude Code、Codex、Copilot、Windsurf、ChatGPT、Lovable、Bolt 或 Replit Agent 生成的代码改动。
已生成 8 条审查项
Run or manually test the path that existed before the AI edit. AI agents can satisfy the new request while quietly breaking old empty states, filters, redirects, or permissions.
What existing user flow could this patch break, and how did you verify it?
Name the command, browser path, API request, or manual checklist that proves the change works. If no automated tests exist, keep the manual gate explicit.
What exact command or manual workflow should fail if this change is wrong?
AI-generated code often handles the happy path but misses loading, empty data, failed requests, double submits, expired sessions, and retries.
Which empty, loading, failed, and retry states does this patch handle?
Check localStorage, cookies, cache keys, feature flags, query params, and persisted preferences so existing users do not get stuck after the AI edit.
Did any persisted state shape, cache key, or cookie contract change?
Search for hardcoded tokens, service keys, webhook secrets, debug flags, and server-only env vars accidentally moved into client code.
Which env vars are required, and are any private values exposed to the browser?
Review request shapes, response fields, status codes, CORS, auth headers, pagination, retries, and how errors surface to the caller.
Which API contract changed, and what existing caller might depend on the old shape?
Check narrow viewport overflow, clipped controls, keyboard focus, labels, button names, contrast, and whether generated UI text fits in its container.
What happens on a 360px wide screen and with keyboard-only navigation?
本工具属于「AI 工作流工具」工作流。 查看完整工作流 · AI 编程项目上线前检查清单
关于此工具
AI 代码审查清单是一个本地 review 工作台，用来审查 Cursor、Claude Code、Codex、GitHub Copilot、Windsurf、ChatGPT、Lovable、Bolt、Replit Agent 等工具生成或修改的代码。
如何使用
选择 AI 工具来源、改动类型、影响范围和审查严格度。补充改动摘要或文件列表后，工具会生成可执行的 review gates。
- 选择 AI 改动来源，以及这是新功能、修 bug、重构、UI、依赖、安全还是部署配置改动。
- 选择影响范围，例如前端、后端、数据库、登录权限、支付、API、多语言、SEO、移动端、测试和文档。
- 按生成的清单审查范围、行为、安全、数据、依赖、移动端、测试、部署假设和交接备注。
- 复制 Markdown review note，放进 PR、issue、代码审查，或下一轮 AI 编程 prompt。
功能特点
这个工具不扫描仓库、不调用 AI 模型。它只根据你提供的上下文，在本地生成审查门禁和后续追问。
- 覆盖 Cursor、Claude Code、Codex、Copilot、Windsurf、ChatGPT、Lovable、Bolt、Replit Agent 等真实 AI 编程工具。
- 根据登录、支付、数据库迁移、API、多语言、SEO、移动端、依赖和部署等影响范围生成条件检查。
- 生成合并、部署或交给下一轮 agent 之前应该追问 AI 的问题。
- 支持勾选、证据备注和 Markdown 导出。
常见问题
AI 生成代码应该怎么 review？
先确认改动文件是否符合原始任务，再检查旧功能是否回归、登录或数据等高风险区域、测试或手动验证门禁，以及部署假设。
这个工具会自动检查我的代码吗？
不会。它是本地 checklist 和 review note 生成器。你选择改动类型和影响范围，工具生成审查项和追问，不上传文件。
非技术 vibe coding 用户能用吗？
能。选择 Lovable / Bolt / Replit Agent 或 ChatGPT 作为来源，然后重点检查改动文件、可见行为、环境变量、部署和具体验证步骤。