工程控制论视角下的软件工程与 Loop Engineering
手上项目一多，人会先崩溃在各种决策和流程控制上，看似是软件工程，实则是软件手工业。尝试从项目管理的分层直觉出发，回到钱学森 1954 年的工程控制论，把反馈回路、层次结构、人在回路三个核心概念对照到 CI/CD、SRE on-call、可观测性、混沌工程和 2026 年流行的 Loop Engineering，最后落回自己正在改造的工作流。
我的实习生 - Hermes 和 Openclaw 中的调研工作流
调研交给 AI，省的是写作时间，没省的是判断成本。这篇用三个量化指标（divergence/coverage/credibility）+ FSM 状态机 + 双层控制回路，把 Wikipedia 漫游时那套自然校正机制重新编码进 Hermes 调研系统。
从 OpenClaw 到 Hermes Agent
龙虾调教系列续篇。前五篇建立在 OpenClaw 上，这篇记录迁移到 Hermes Agent 之后，如何把原来那套改造思路在新底座上重新实现——以及哪些地方变得更好了。
龙虾调教（完结篇）：从零搭建项目管理的 OpenClaw 助理
前四篇文章解释了为什么这样设计。这篇只讲怎么做——从零开始，把 OpenClaw 配置成一个帮你跑 Linear + GitHub 全链路项目管理的 AI 助理。
龙虾调教（四）：项目管理工程实践优化
把 OpenClaw 从「流程可用」推到生产：持久记忆、脚本执行层、端到端项目自动化。逐项拆解每个改动背后的判断。
龙虾调教（三）：OpenClaw 软件产品开发的项目管理工作流
本文分享如何以 Linear + GitHub 为管理底座，让 OpenClaw 担任敏捷教练和项目管理角色，覆盖需求收集、Sprint 规划、Code Review、验收到回顾的完整开发节奏，让人只在关键决策点介入，其余监控与自动化全部交给 AI。
龙虾调教（二）：让 OpenClaw 帮你把重复工作变成 SOP
本文介绍如何在 OpenClaw 中构建三层 Workflow 系统——通过 AGENTS.md 元规则、WORKFLOW.md 索引和 Playbook 文件，将重复性工作 SOP 化，让 AI 助手读完几个文件就能独立执行任何重复任务，无需每次重新解释。
龙虾调教（一）：成长为个人助手
本文记录了从默认安装出发，逐步将 OpenClaw 配置为私人助理的完整过程——注入身份与行为规范、建立跨 session 的持久记忆、扩展自定义 Skills的过程，让它真正融入你的工作流，而不只是一个问答工具。
别再对着 AI 干瞪眼了：常用 Prompt 框架整理
整理了十几个常用的 Prompt 结构化框架，从 ROSES 到 CO-STAR，帮你把和 AI 的对话从闲聊变成协作
来自 2025年的 Hello World
如何在2025年使用 AI 从0到1构建一个网站
Software Engineering and Loop Engineering Through the Lens of Engineering Cybernetics
Juggle enough projects and you crash first on decisions and process control — it looks like software engineering, but it is software craftsmanship. Starting from the layered intuition of project management, this post returns to Qian Xuesen's 1954 Engineering Cybernetics, maps feedback loops, hierarchy, and human-in-the-loop onto CI/CD, SRE on-call, observability, chaos engineering, and 2026's Loop Engineering, and lands back on a workflow I am rebuilding.
My Intern — Research Workflows in Hermes and OpenClaw
Hand research to AI and you save writing time — not judgment cost. This post re-encodes the natural correction loop from Wikipedia wandering into the Hermes research system using three metrics (divergence/coverage/credibility), an FSM, and a dual-layer control loop.
From OpenClaw to Hermes Agent
The continuation of the OpenClaw Training series. The first five posts were built on OpenClaw; this one documents how to reimplement that same customization approach on Hermes Agent — and where things got better.
OpenClaw Training (Final): Building a Project Management AI Assistant from Scratch
The first four posts explained the design rationale. This one is purely about how to do it — configuring OpenClaw from scratch into an AI assistant that runs your full Linear + GitHub project management pipeline.
OpenClaw Training (IV): Production-Grade Engineering Upgrades
Pushing OpenClaw from 'functionally usable' to production-ready: persistent memory, a script execution layer, and end-to-end project automation. A breakdown of every change and the reasoning behind it.
OpenClaw Training (III): A Project Management Workflow for Software Teams
How to use Linear + GitHub as the management foundation while letting OpenClaw serve as agile coach and project manager — covering the full development cycle from requirement intake to sprint retrospectives, with humans only at key decision points.
OpenClaw Training(II): Using OpenClaw to Turn Repetitive Work into SOPs
How to build a three-layer Workflow system in OpenClaw — using AGENTS.md meta-rules, a WORKFLOW.md index, and Playbook files — to turn repetitive tasks into SOPs that an AI assistant can execute independently after reading just a few files.
OpenClaw Training (I): Growing into a Personal Assistant
A complete walkthrough of configuring OpenClaw from its default state into a personal assistant with identity, persistent memory, and proactive capabilities
Stop Staring Blankly at AI: A Guide to Common Prompt Frameworks
A roundup of over a dozen structured Prompt frameworks — from ROSES to CO-STAR — to turn your AI conversations from small talk into real collaboration
Hello World from 2025
How to build a website from scratch using AI in 2025