Codex 工程变更与审查工作流
Rootloom
v4.4.0 / 已发布
面向 OpenAI Codex 的本地工程工作流，把风险、根因、变更范围与实际验证组织成可检查的工程过程。
OpenAI Codex · Codex Plugin · Python 3.11+ · AGENTS.md · Git
Yanqing Li: Careful work,and still curious.
I do Senior Test Development Engineer at Autohome (汽车之家), and have spent 14 years in quality engineering; alongside that I keep building AI tools, open-source projects and my own products.
I tend to turn a problem I keep running into into a tool: PulseGuard checks frontend and API liveness across several projects at the company; CaseFlux draws requirements, cases, real browser runs and failure diagnosis into one line; the open-source Rootloom and 衡木, which make an agent’s changes and architecture decisions inspectable.
Next I want to go deeper into AI testing and development, and to keep writing the process down.
Work
In-house tools shipped inside the company, products still in development, and open-source projects. Each one states its real status: published, private demo, or still changing.
All 10 projects-
Rootloom 概念图 -
CaseFlux flow diagram · drawn from the AI testing loop in the résumé, not a product screenshot AI 原生测试自动化平台
CaseFlux
私有演示 / 持续迭代
个人持续研发的 AI 测试平台，连接需求、版本化用例、真实浏览器执行、失败证据与修复审核，可面试演示。
需求解析 · 功能测试用例 · UI 自动化 · Playwright · 执行证据
-
PulseGuard 项目封面：内网 UI / API 探活工作台 UI/API 探活与质量监测
PulseGuard
公司多个项目实际应用
自研并应用于公司多个项目的前端与 API 探活。统一管理健康状态、失败证据、告警与跨网络执行节点，支持公共脚本库和 MCP 诊断。
FastAPI · SQLite WAL · Playwright Python · React · TypeScript
-
Cognera woven label · no public screenshot 个人智能工作台
Cognera
私有项目 / 持续迭代
Server-first 的个人智能工作台，覆盖 Web、iOS 与 macOS Cowork，并把来源、理解、Skills、连接和权限边界纳入统一产品模型。
Python · React · SwiftUI · Tauri · OpenAPI
-
衡木项目封面 架构审查与目标设计
衡木
v1.2.1 / 已发布
从仓库事实或设计目标出发，连接架构审查、独立核实、目标架构、技术决策、改造计划与确定性质量门禁。
Python 3.11-3.14 · JSON Schema · YAML · Git · GitHub Actions
-
AI Playwright Framework：YAML 测试资产分层 AI 自动化测试框架
AI Playwright Framework
开源 Alpha / 持续迭代
以 YAML 管理可审查的 Playwright / pytest 测试资产，保留确定性执行，并支持经浏览器验证的 AI 生成、定位恢复和自然语言任务。
Python · Playwright · pytest · YAML · Allure
Also in the workshop
Writing
Pattern is what the character 文 meant first: the figure woven into cloth. These are the marks the work leaves — AI testing, agent engineering, and the judgements and trade-offs in between.
All 6 articles
从对话驱动到状态驱动：用 Notion 让人和 AI 接力工作
AI 协作真正缺少的，往往不是更强的模型，而是一份不会随着聊天结束而消失、能被人和不同 AI 持续维护的当前工作状态。
Read this- Loop Engineering：为 Coding Agent 设计可控的反馈循环 方法论
- AI 产品团队做测试左移，先把标准说清楚 方法论
- LLM 输出不稳定时，自动化测试应该怎么断言 方法论
- AI 产品自动化测试框架怎么拆：Pytest + Playwright + Allure 实践
RSS feedNew articles appear here first
Career
The warp goes on the loom before the weft has anywhere to pass. 14 years of testing and quality engineering — web, apps, mobile web, mini programs, extensions and admin systems — is the ground under everything else.
Full résumé- Earlier careerTest Engineer | Renren, Lashou, Daojia, Shouqi2012.04 — 2017.10
- MockingBot (墨刀)Test Engineer / Test Development Engineer2017.11 — 2021.01
- Lanhu (蓝湖)Test Development Engineer / Test Lead2021.02 — 2024.08
Now · Autohome (汽车之家) · Senior Test Development Engineer
- Own quality assurance for AI marketing tools, the Douyin livestreaming assistant, direct-message and danmaku auto-reply, the car marketplace, H5, mini-program and back-office systems, covering functional, API, UI, exception-path, risk-control and stability verification.
- Lead a testing team of about 5, running requirements review, test planning, task breakdown and release gating, and working with roughly 20 engineering, product and business colleagues to close out quality risks and defects.
- Built and maintain 100+ UI automation cases and dozens of API cases, reaching 90%+ automation coverage of core business within my scope; API regression runs in CI/CD and has caught problems at build time.
Ideas keep growing in conversation.
About a project, a quality practice, or AI testing and development — I am glad to hear from you. 结 is a knot, and also to make someone’s acquaintance.
lyqgmail@gmail.com