Motto: “Stay hungry, stay foolish” by Steve Jobs.
我是 Misty-Star / Zhengyu Chen，一个偏 AI / LLM / Agent / Backend 方向的 AI Native Coder。
I build software with LLMs, coding agents, backend systems and automation-first workflows.
我不只是“用 AI 辅助写代码”，更关注如何把 Codex、Claude Code、Cursor、MCP、RAG、Tool-use、Function-call、Agent Skills 等能力组织成可重复、可验证、可演进的工程系统。
目前我是一名 Codex / Claude Code 重度使用者，长期跟进 LLM / Agent 浪潮，并持续探索人类开发者与 AI Agent 协同编程的新范式。
White base. EVA-inspired purple / green / orange accents.
白底为主，带一点 EVA 风格的紫、绿、橙信号色。
2025 Mid ── Started programming heavily with Claude Code & Cursor
2025 Late ── Shifted into a Codex-heavy workflow
Now ── Building agentic coding workflows and engineering loops一路从：
- Prompt Engineering — 写出更清晰的目标、约束与指令
- Context Engineering — 组织更好的上下文、记忆、检索与仓库 grounding
- Harness Engineering — 为 Agent 设计工具、MCP、脚手架与执行环境
- Loop Engineering — 构建计划、执行、验证、反馈、迭代的自动化闭环
From prompting better models to designing better agent loops.
- AI Native Coding Workflow — Codex-first / Claude Code-assisted development
- Agentic Software Engineering — planning, execution, verification and feedback loops
- LLM Application Backend — RAG, workflow orchestration, API design, service integration
- Tool-use & Function Calling — turning models into operators with reliable tools
- MCP / A2A / Agent Skills — connecting agents, tools, contexts and capabilities
- Low-code AI Workflows — FastGPT, Dify, Coze and practical automation pipelines
| 项目 | 描述 | 技术 | 状态 |
|---|---|---|---|
| 智卷云 | 基于LLM的AI智能组卷平台 | NestJS + Keycloak + RAG + ReAct Agent + Puppeteer | ✅ 已完成 |
| Obsidian-Ultimate-Publisher | Obsidian 文章发布插件 | JavaScript + TypeScript | 🔥 进行中 |
| grok-it-mcp | 链接Grok订阅能力的Agent Plugin | Codex + Claude Code + Skills + MCP | ✅ 已完成 |
| AI Coding Workflow | 设计更好的人类×Agent协作闭环 | Claude Code + Codex + MCP + Agent Loop | 🔥 持续进化 |
I believe the next generation of software engineering is not only about writing code faster,
but about designing better human-agent collaboration loops.
我相信下一代软件工程的核心不只是“更快地写代码”，而是设计更好的 人类 × Agent 协作系统。
Good agents need:
- clear goals / 清晰目标
- rich context / 充分上下文
- reliable tools / 可靠工具
- verifiable feedback / 可验证反馈
- repeatable loops / 可复用闭环
GOAL → CONTEXT → TOOLING → EXECUTION → VERIFICATION → LOOP
目标 上下文 工具化 执行 验证 闭环A little bit of anime, a little bit of backend, a lot of agents.
喜欢白底、克制、干净的页面，也喜欢带一点 EVA-like signal color 的视觉风格：
WHITE_BASE #FFFFFF
EVA_PURPLE #6B3FA0
SIGNAL_GREEN #8CE10B
WARNING_ORANGE #FF7A00
DARK_TEXT #1F1F1F- Blog: https://mistystar.cn/
- GitHub: @Misty-Star
