跨域笔记
知识中心
关于我构建、写作或交易过的系统的笔记 — 按领域分类。资历与已发行作品见首页；这里是思考所在。
智能体架构
从 Cursor 入门到仓库级 AGENTS.md、私有知识库、MCP 工具设计、RL 训练模型与单智能体编码循环的精选路径。
- Cursor 入门：自动化整个循环 Treat Cursor as an automation surface — rules, tests, MCP, and agents run the boring parts so you keep one thread for judgment. 阅读 →
- 面向 AI 编码代理的 AGENTS.md 标准 One open Markdown file per repo — closest AGENTS.md wins; user prompts override. 阅读 →
- 为 AI 代理组织知识 Pair a public site with a private brain — progressive disclosure, lifespans, verification metadata. 阅读 →
- 用 search 与 execute 构建 MCP 服务器 Two MCP tools — search and execute — beat hundreds of per-operation tool definitions. 阅读 →
- 强化学习提升代理模型中的工具调用能力 RL specializes models for the tool loop — Composer is the public case study. 阅读 →
- Keep the thread: why plan–critique–build pipelines fail modern coding agents One agent, one thread, small units — verify with tests, not lossy handoffs. 阅读 →