ANTHROPC - Claude Code & Agent SDK

参考 ANTHROP\C - Claude > ANTHROPC - Claude Opus/Sonnet 5 & Fable 5 & Mythos 5 & AI-Coding & Text-to-Code: SQL > Script > DSL 2022-2025；
对标 GitHub Copilot & OpenAI Codex - OpenAI Codex CLI/APP & AGENTS.md & Agents SDK / AgentKit & Cognition Labs - Devin - DeepWiki & Google - AI Studio & Gemini CLI > AntiGravity & Agents-CLI；

20250225 Claude 3.7 Sonnet and Claude Code > Claude Code - Deep coding at terminal velocity > Claude Code 是首个真正意义上的 Devin；> Claude Code overview & Set up Claude Code > 0715 安装 Windows 版：先行使用 ANTHROPIC_API_KEY（及其消耗 Token），后配置 Kimi-K2 & QWen3-Coder 等；另，VS Code & Cursor 中安装官方插件配合使用；> 0910 AnyRouter 这类“中转站”泛滥（Claude Code 的普及）；> 拟 1008 开通 Pro 计划使用；> 1024 近期拟在 WSL 容器中安装 Claude Code 等在 Windows 下运行不稳定的 CLI；> 1101 使用官方 Windows 版安装方案（不依赖 Node 容器而为独立的 Claude，2.0.31）：irm https://claude.ai/install.ps1 | iex； 可执行 claude update 手工更新；

• Slash commands - Control Claude's behavior during an interactive session with slash commands. > 让 Claude Code 超深度思考（ultrathink）的自定义指令
• Hooks - Customize and extend Claude Code's behavior by registering shell commands
• Model Context Protocol (MCP) - Connect Claude Code to tools via MCP > 参考 Model Context Protocol
• Sub agents - Create and use specialized AI sub agents in Claude Code for task-specific workflows and improved context management. > Subagents 不要用来执行任务，而是用来查找信息，做调研任务，然后输出结论给主线程，指导主线程的下一步行动。
• Output styles - Adapt Claude Code for uses beyond software engineering - Default / Explanatory / Learning
• Checkpointing - Automatically track and rewind Claude's edits to quickly recover from unwanted changes.
• Headless mode - Run Claude Code programmatically without interactive UI > CLI reference - Complete reference for Claude Code command-line interface, including commands and flags.

Claude Code 202506 开始 Pro 账号也可使用 > 0606 Claude Code + Opus 4 = The Future of AI Coding Agents & 0607 @宝玉 xp - 整理一些我使用 CluadeCode 的经验和心得 - 敏捷设计 + 代码生成 & @hylarucoder - Claude Code 试了几天，最适合的后端语言是 Go；> 前段时间还到处推荐 TaskMaster.AI，现在感觉就是 ClaudeCode 的一个丐版，Claude-3.7-Sonnet 主模型，PPLX 的 Sonar-Pro 调研模型。如同之前 o3 及其 DeepResearch 的能力代差 & 高费用，Model/Token 在软工的价值评估体系中的占比会不断提升（算 ROI 还是便宜），某种意义上“工程”的理想也指日可待了。

Claude Code in Action - Practical walkthrough of using Claude Code to accelerate your development workflow

• anthropics / claude-code-action > cholf5 / claude-code-in-action: 《Claude Code in Action》中文版
• anthropics / claude-quickstarts: A collection of projects designed to help developers quickly get started with building deployable applications using the Claude API
• anthropics / claude-cookbooks: A collection of notebooks/recipes showcasing some fun and effective ways of using Claude.
• luongnv89 / claude-howto: A visual, example-driven guide to Claude Code — from basic concepts to advanced agents, with copy-paste templates that bring immediate value. > Claude Code How-To Guide - Master Claude Code in a Weekend
• zebbern / claude-code-guide: Here is a full guide on claude tips and tricks and how you can optimise your claude code the best full guides and every command possible even hidden ones!
• Cranot / claude-code-guide: Claude Code Comprehensive Guide
• peterkrueck / Claude-Code-Development-Kit: Claude Code AI Development Framework > Handle context at scale - my custom Claude Code workflow including hooks, mcp and sub agents
• Claude Code 中文教程 - Claude Code 极速入门：重塑你的编程工作流
• hesreallyhim / awesome-claude-code: A curated list of awesome commands, files, and workflows for Claude Code

Workflows & Knowledge Guides & Tooling (IDE Integrations) & Hooks & Slash-Commands (Version Control & Git & Code Analysis & Testing & Context Loading & Priming & Documentation & Changelogs & CI / Deployment & Project & Task Management & Miscellaneous）& CLAUDE.md Files（Language-Specific & Domain-Specific &  Project Scaffolding & MCP）& Official Documentation

0418 Claude Code: Best practices for agentic coding > 1105 Min Choi@minchoi: This Claude Code prompt literally turns Claude Code into ultrathink visionary🤯
0527 Claude Researcher Explains How Agentic Search Performed Better Than RAG For Code Generation
0725 How Anthropic teams use Claude Code > Anthropic 团队如何玩转 Claude Code - How Anthropic teams use Claude Code
0824 Claude Code 究竟牛在哪里？（以及如何在你的 AI 智能体中复刻它的魔法！）- What makes Claude Code so damn good (and how to recreate that magic in your agent)!?
0911 为 AI 智能体打造高效工具——让 AI 智能体来帮忙 - Writing effective tools for agents — with agents
0929 Effective context engineering for AI agents

1020 Claude Code on the web > 对标 Codex Cloud；

1126 打造高效框架，让 AI 智能体胜任“长跑”任务 - Effective harnesses for long-running agents > 初始化智能体(Initializer Agent)：最初的第一个会话使用专门的提示词，要求模型设置初始环境：一个 init.sh 脚本，一个记录智能体操作日志的 claude-progress.txt 文件，以及一个展示添加了哪些文件的初始 Git 提交。编程智能体(Coding Agent)： 后续的每一个会话都要求模型取得一点增量进展，然后留下结构化的更新记录。> 相同策略参考 Specification-Driven Development(SDD) & Spec-Coding - Fission-AI / OpenSpec；

1203 How AI is transforming work at Anthropic > AI 如何改变 Anthropic 的工作方式 > 本文聚焦于人工智能对工作方式的改变，以 Anthropic 公司为例展开研究。通过对 132 名工程师和研究人员的调查、53 次深度定性访谈及 Claude Code 使用数据的分析，发现 AI 使用正给软件开发人员的工作性质带来根本性变化，既带来希望也引发担忧。> 以 Claude Code 发布为节点：之前，技术团队是资产；之后，不会 Vibe Coding 的技术团队是负债。> 传统的技术团队会逐渐收缩，变成基础设施团队；至于实现产品功能，未来可能是一种新的组织形式。

20260106 Claude Code on desktop - Run Claude Code tasks locally or on secure cloud infrastructure with the Claude desktop app

0130 How AI assistance impacts the formation of coding skills - Coding with AI led to a decrease in mastery—but this depended on how people used it. > “AI 让我更快”和“AI 帮我学到东西”之间的差距，比想象中大太多。大部分人的工作流，只优化了前者，却在悄悄牺牲后者。

0110 Merged /commands and skills in 2.1.3 update > 自然语言驱动的第一性原理；

0206 Orchestrate teams of Claude Code sessions - Coordinate multiple Claude Code instances working together as a team, with shared tasks, inter-agent messaging, and centralized management. > 设置 "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1" 启用，区分 Agent Teams & Agent Swarm vs Sub-Agent；

0206 Building a C compiler with a team of parallel Claudes - We tasked Opus 4.6 using agent teams to build a C Compiler, and then (mostly) walked away. Here's what it taught us about the future of autonomous software development. & 0223 How AI helps break the cost barrier to COBOL modernization - Legacy code modernization stalled for years because understanding legacy code cost more than rewriting it. AI flips that equation.

0408 Scaling Managed Agents: Decoupling the brain from the hands - Harnesses encode assumptions that go stale as models improve. Managed Agents—our hosted service for long-horizon agent work—is built around interfaces that stay stable as harnesses change.

openma - The open-source Claude Managed Agents alternative - Run Claude agents yourself — on Cloudflare or with docker compose up - A wire-compatible, self-hostable agent platform: durable sessions, sandboxed code execution, vault-backed credentials, MCP server hosting, and Claude Code skills — served from Cloudflare Workers + Durable Objects, or a Docker container on your own box. Same API surface as the proprietary version, with the source code and the runtime under your control.

• open-ma / open-managed-agents: openma - open-source, self-hosted implementation of Claude's Managed Agents API. Drop-in compatible. Runs on Cloudflare Workers + Durable Objects or Node.js. Apache 2.0.

0514 How Claude Code works in large codebases: Best practices and where to start - The most successful Claude Code deployments share a set of recognizable patterns across configurations, tooling, and org structure. This article is part of Claude Code at scale, a new series covering best practices for engineering organizations building with Claude Code at enterprise scale. > Claude Code 在大型代码库中的运作方式：最佳实践与入门指南

0528 Introducing dynamic workflows in Claude Code > /effort 选择 ultracode(xhigh+workflows)，然后 /goal；

20240506 Companies Are Just a Graph of Algorithms - AI is about to see your company as a series of components to be optimized & 20240924 Policy, SOPs, and AI Are All You Need - How AI-powered Policy and SOPs will soon run teams, departments, and companies
Cursor 的 YOLO > OpenSpec 的 SDD > Ralph Loop > Karpathy 的 AutoResearch > Claude Code 的 Plan mode > Codex 的 /goal > Claude Code 的 ultracode（dynamic workflow）> Ralph loop 是个分水岭，之前模型迭代竞争，之后是模型 + Harness 竞争；

0603 How Anthropic enables self-service data analytics with Claude

传统方式下，依靠宽表、非标准化数据表降低非技术人员数据使用门槛，会随业务扩张出现视图重叠、定义混乱的问题；而搭建封闭使用环境又会遗漏长尾业务问题，造成指标和仪表盘冗余、团队数据工作割裂。大语言模型虽为自助分析提供了新方向，但直接让模型对接数据仓库执行任务，容易产生虚假精准感，还会割裂使用者与底层数据基础设施、专业文档及数据知识的联系。……数据分析代理出错的核心并非代码生成问题，而是上下文匹配与信息验证问题，并总结出概念与实体模糊、数据陈旧、信息检索失败三大主要错误诱因。为此，团队搭建了分层式代理分析堆栈，包含数据基础、权威信息源、技能体系三大核心层级，同时配套离线评估、消融实验、线上验证等多维度效果检测手段，还给出了技能文档模板、从零落地的实施思路，以及实践中踩过的无效方案与对应的经验教训，为各类数据团队使用大模型搭建智能分析体系提供了全套最佳实践参考。
第一层数据基础层依托规范的数据模型、数据测试、元数据管理等工程实践，打造唯一权威的标准数据集，淘汰重复近似的数据表，从源头减少概念与实体模糊问题，同时借助统一代码仓库和流程约束，延缓数据陈旧问题；第二层权威信息源层包含语义层、数据血缘、历史查询库、业务知识图谱，帮助模型将用户自然语言问题转化为标准数据实体，进一步厘清数据含义；第三层技能层定义模型的操作流程、检索规则与分析范式，专门解决检索失败问题，三层架构相互配合，大幅提升分析准确性。

0618 Claude Code now supports artifacts - Preview your in-progress work as a live, interactive web page—built from your full session context and shareable with your team. > 对标 Codex 的 Sites；

20260626 近期 Claude Code 中 GLM-5.X 接入报 API 错误，故采用 DeepSeek-v4 系列接入使用；与 Claude for Desktop 的配置各自独立；

Plugins - Extend Claude Code with custom commands, agents, hooks, and MCP servers through the plugin system. > Plugin marketplaces - Create and manage plugin marketplaces to distribute Claude Code extensions across teams and communities.

基于 Claude Code 增强的第三方中，启动较早的 SuperClaude-Org / SuperClaude_Framework 这类，多是在用户 .claude 根目录以及 commands/agents/hooks 等子目录中直接挂提示配置，而一些体量较小第三方则在项目 .claude 目录中小范围配置，各有利弊。等到 Plugins & Markplaces 方案出来后，插件配置体系就清晰多了，比如 davila7 / claude-code-templates 就更能按需所取。从这个角度，后来 Agent Skills 的特别之处，就是通过固化的代码/伪代码，平衡了提示等资产过于侧重文本，导致的映射跃层不确定与 Token 消耗，也是 AI-Coding 第一性的体现。另外，大家都会跟进这个方案，比如在 Gemini CLI/Github Copilot/Cursor 等中跑 github / spec-kit。

• anthropics / claude-plugins-official: Anthropic-managed directory of high quality Claude Code Plugins.

Agent Skills & Agent Skills - Create, manage, and share Skills to extend Claude's capabilities in Claude Code. > 1016 Equipping agents for the real world with Agent Skills > 尤其关注 Meta Skills - skill-creator & template-skill > 详见 ANTHROPC - Claude Code - Plugins & Agent Skills & Agent SDK；

• anthropics / skills: Public repository for Skills > /plugin marketplace add anthropics/skills 添加 Marketplace 后安装 Plugin，Git 连接需通畅；
• travisvn / awesome-claude-skills: A curated list of awesome Claude Skills, resources, and tools for customizing Claude AI workflows — particularly Claude Code

Claude Code SDK - Build custom AI agents with the Claude Code SDK > 20250930 Building agents with the Claude Agent SDK > Migrate to Agent SDK overview - Build custom AI agents with the Claude Agent SDK > 支撑 GitHub Actions & GitLab CI/CD > 详见 ANTHROPC - Claude Code - Plugins & Agent Skills & Agent SDK；

Agent 开发应该是到了临界点。使用类似于 Langchain 的 Deep Agents、Claude 的 Agents SDK 这种自带工具（Bash、Web、Todo、Subagent 等）的 Agent 框架，然后结合 UI，可以方便的打造多轮长时智能体。需要一套Agent UI+ Sandbox runtime。

官方模块：

• anthropics / claude-code: Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.

morso233 / claude-code：Claude Code 不是开源的，不过早期版本（v0.29-）发布的时候，连着 SourceMap 一起发布的，所以有人根据 SourceMap 逆向了原始代码；
shareAI-lab / analysis_claude_code: 本仓库包含对 Claude Code v1.0.33 进行逆向工程的完整研究和分析资料。包括对混淆源代码的深度技术分析、系统架构文档，以及重构 Claude Code agent 系统的实现蓝图。主要发现包括实时 Steering 机制、多 Agent 架构、智能上下文管理和工具执行管道。该项目为理解现代 AI agent 系统设计和实现提供技术参考。> 文档主要由 AI 生成。
Claude Code: An analysis: This entire report was generated by Claude Opus 4, with the help of almost all of the major flagship models. However, the 8000 word write-up on the process of making this report was manually written - you can always start there: Conducting smarter intelligences than me: new orchestras

• anthropic-experimental / sandbox-runtime: A lightweight sandboxing tool for enforcing filesystem and network restrictions on arbitrary processes at the OS level, without requiring a container. > 利用了操作系统自带的沙箱机制（macOS 用 sandbox-exec，Linux 用 bubblewrap），并结合代理网络过滤，能安全地限制代理、MCP 本地服务器、bash 命令及任意进程的行为。
• anthropics / claude-code-sdk-python > 参考上述 Claude Code SDK，另可 Add Claude Code to your IDE
• Claude Code for VSCode: Claude Code for VS Code: Harness the power of Claude Code without leaving your IDE > 与 Windows 版共享配置与环境，且同步迭代更新；更明确的 Slash Commands；
• Run parallel Claude Code sessions with Git worktrees > Branch 管“路线”，Worktree 管一个“路线”的多个并行“现场”。全矩阵的 Peter Steinberger 模式：多条分支，每条分支再开多个 Worktree，甚至分发到多台机器。

模板插件及反身性教学：

• UfoMiao / zcf: Zero-Config Code Flow for Claude code & Codex > 可快速入门
• davila7 / claude-code-templates: CLI tool for configuring and monitoring Claude Code

aitmpl: agents & commands & settings & hooks & mcps > plugins & skills，前期调试参见下文，后续调试详见 ANTHROPC - Claude Code - Plugins & Agent Skills & Agent SDK；> Daniel San@dani_avila7: My Claude Code Learning Path

• wshobson / agents: Intelligent automation and multi-agent orchestration for Claude Code > Claude Code Plugins: Orchestration and Automation

• carlvellotti / claude-code-pm-course: Interactive course teaching Product Managers how to use Claude Code effectively > Claude Code for Product Managers

• ChrisWiles / claude-code-showcase: Comprehensive Claude Code project configuration example with hooks, skills, agents, commands, and GitHub Actions workflows

• diet103 / claude-code-infrastructure-showcase: Examples of my Claude Code infrastructure with skill auto-activation, hooks, and agents > Claude Code is a Beast – Tips from 6 Months of Hardcore Use

• musistudio / claude-code-router: Use Claude Code as the foundation for coding infrastructure, allowing you to decide how to interact with the model while enjoying updates from Anthropic.

• GAC Claude Code 镜像官网 国内团队账号池共享的 Claude Code 方案（后续大量中转站）> 20250707 GMail 注册登录（6siZF4aH!xry6JS）；

• https://apic1.ohmycdn.com/api/v1/ai/openai/cc-omg/v1/messages 试用链接，有赠送额度；

第三方扩展：随着 Claude Code 的功能迭代，会覆盖大量第三方的扩展定位，重点关注落地 ** Context Engineering & Agentic Engineering > Harness Engineering & Loop Engineering ** > 0908 Claude 代码框架之战

• coleam00 / context-engineering-intro: Context engineering is the new vibe coding - it's the way to actually make AI coding assistants work. Claude Code is the best for this so that's what this repo is centered around, but you can apply this strategy with any AI coding assistant! > GitYCC / context-engineering-intro-zh: Context engineering 是新的 Vibe Coding —— 它是讓 AI 程式助理真正發揮作用的關鍵方式。Claude Code 是目前最適合做這件事的工具，所以這個 repo 會以它為核心，但其實你也可以把這個策略應用在任何 AI 程式助理上！

• NomenAK / SuperClaude: Meet SuperClaude – The Missing Power-Up for Claude Code - A configuration framework that enhances Claude Code with specialized commands, cognitive personas, and development methodologies. > SuperClaude-Org / SuperClaude_Framework: A configuration framework that enhances Claude Code with specialized commands, cognitive personas, and development methodologies. - SuperClaude is a meta-programming configuration framework that transforms Claude Code into a structured development platform through behavioral instruction injection and component orchestration. 后续详见 SuperClaude Framework & SuperPowers；

• Pimzino / claude-code-spec-workflow: Automated workflows for Claude Code. Features spec-driven development for new features (Requirements → Design → Tasks → Implementation) and streamlined bug fix workflow for quick issue resolution (Report → Analyze → Fix → Verify).

• OneRedOak / claude-code-workflows: The best workflows and configurations I've developed, having heavily used Claude Code since the day of it's release. Workflows are based off applied learnings from our AI-native startup.

• breaking-brake / cc-wf-studio: ClaudeCode Workflow Studio

• BloopAI / vibe-kanban: Kanban board to manage your AI coding agents > Vibe Kanban

• Maciek-roboblog / Claude-Code-Usage-Monitor: Real-time Claude Code usage monitor with predictions and warnings

• opactorai / Claudable: Claudable is an open-source web builder that leverages local CLI agents, such as Claude Code, Codex, Gemini CLI, Qwen Code, and Cursor Agent, to build and deploy products effortlessly.

• BeehiveInnovations / zen-mcp-server: The power of Claude Code + [Gemini / OpenAI / Grok / OpenRouter / Ollama / Custom Model / All Of The Above] working as one.

• getAsterisk / claudia: A powerful GUI app and Toolkit for Claude Code - Create custom agents, manage interactive Claude Code sessions, run secure background agents, and more. > winfunc / opcode: A powerful GUI app and Toolkit for Claude Code - Create custom agents, manage interactive Claude Code sessions, run secure background agents, and more. > opcode

• CC Mate - Configure your Claude Code without pain - Manage your Claude Code on a beautiful GUI

• jarrodwatts / claude-hud: A Claude Code plugin that shows what's happening - context usage, active tools, running agents, and todo progress

• MarkShawn2020 / lovcode - A desktop companion app for AI coding tools. Browse Claude Code chat history, manage configurations, commands, skills, and more.

• ruvnet / claude-flow: Claude-Flow v2.0.0 Alpha represents a revolutionary leap in AI-powered development orchestration. Built from the ground up with enterprise-grade architecture, advanced swarm intelligence, and seamless Claude Code integration. > 🌊 The leading agent orchestration platform for Claude. Deploy intelligent multi-agent swarms, coordinate autonomous workflows, and build conversational AI systems. Features enterprise-grade architecture, distributed swarm intelligence, RAG integration, and native Claude Code support via MCP protocol. Ranked #1 in agent-based frameworks

• automazeio / ccpm: Project management system for Claude Code using GitHub Issues and Git worktrees for parallel agent execution. > Automaze - Technical Co‑Founder & CTO as a Service - The zero-equity way for founders to bring their idea to life, attract users, and achieve product-market fit.

• CONDUCTOR - Run a team of coding agents on your Mac. - Create parallel Codex + Claude Code agents in isolated workspaces. See at a glance what they're working on, then review and merge their changes.

• smtg-ai / claude-squad: Claude Squad - Manage multiple AI agents like Claude Code, Codex, and Aider. - 10x your productivity

• stravu / crystal: Run multiple Claude Code AI sessions in parallel git worktrees. Test, compare approaches & manage AI-assisted development workflows in one desktop app.

• steipete / claude-code-mcp: Claude Code as one-shot MCP server to have an agent in your agent.

• YishenTu / claudian: An Obsidian plugin that embeds Claude Code as a sidebar chat interface

开源及商业模式：

• opencode-ai / opencode: opencode - A powerful AI coding agent. Built for the terminal. > 号称 Claude Code 开源实现；> charmbracelet / crush: The glamourous AI coding agent for your favourite terminal 💘 > 后续详见 OpenCode - Zen / OMO(LazyCodex) & Crush；
• plandex-ai / plandex: Open source AI coding agent. Designed for large projects and real world tasks. > plandex - Build real world software with AI. - An open source, terminal-based AI coding agent designed to handle large projects, large files, and large tasks. Featuring a diff review sandbox, full auto mode, and smart context management up to 2M tokens.
• Use Claude Code without paying anything - Giga AI Free partners with the top AI companies to provide free Claude Code supported by ads and data training.

• oraios / serena: A powerful coding agent toolkit providing semantic retrieval and editing capabilities (MCP server & other integrations) > 基于 LSP (Language Server Protocol) 提供符号级检索，即通过函数名、变量等的定义、引用和依赖关系，预解析整个代码库的结构；> 202607 调试 Cursor - Composer & Cloud Agent & SDK - 2026，确认不再建议使用 MCP 或 Plugin Marketplace 模式，而通过命令行安装与初始化

ECC Tools - OPEN AGENT HARNESS SYSTEM - Skills, agents, and security for your coding agent.- Pick a profile. Install the skills and agents your team needs. AgentShield scans every session. The GitHub App turns repo history into reusable defaults. 140K+ stars on GitHub. Start free, scale to enterprise.

• affaan-m / ECC: The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond. > 来自 Anthropic 黑客马拉松获胜者的完整 Claude Code 配置集合，即把 AI 编程工具当成基础设施来用的一整套完整系统。

20250623 @blackanger: 大概 50万行代码规模的项目，用 claude code 生成的代码质量远远不如 cline+gemini 2.5 pro。要达到同等质量，使用 claude code 必须要人盯着。然而 cline + gemini 2.5 pro 的问题是太贵了。> 参考 Claude Dev > Cline / Roo Code & Augment Code & Kiro & Kilo & Tessl Cofounder；

0625 @zbeyens: Building an AI editor with ... & 0622 @axtrur: 高强度使用 claude code 后，100刀/月的 max 还是太贵，折腾体验了 5种方案 - Proxy / MCP with other model / Reverse API / Account Share / Manual Plan & 0629 刘小排是如何使用 Claude Code 的？ > 0714 @刘小排：完全合法、无需网络、便宜使用 Claude Code 的方法 & 0715 @trq212: Claude Code is All You Need & 0721 @RayFernando1337: Tell Claude Code to use sub-agents for lengthy prompts and huge bugs and it'll start to cook in parallel.

0701 I Shipped a macOS App Built Entirely by Claude Code > 一位资深 macOS 开发者分享了他的亲身经历：用 Claude Code + SwiftUI，打造了一款完整的调试工具 Context，全程几乎不写代码。……未来的 IDE，不会再以“代码编辑”为中心，而是以“上下文管理 + 代理执行”为核心。

• indragiek / Context: Native macOS client for Model Context Protocol (MCP)

0703 @Yangyixxxx 推荐 aibuilderclub: AI 写好 PRD > TaskManager 切分任务 > 用 Claude Code 定义3个 UI 设计师 > git worktree 切开以便不影响原有内容 > 开始出设计，三个方案你总能选一个进行优化 > 大差不离之后，安个 Stagewise 插件就可以针对性选择一些 Component 进行精确性调整；运用好这些 AI 工具，在当下决定了普通程序员和 AI 程序员的效率差别可能是 10倍甚至百倍的，或许在未来 2-3年会被更傻瓜的工具抹平，但现在是有明显的技术红利期。& 0703 My Claude Code Workflow And Personal Tips - How I use roadmap + task files to manage Claude Code, and my personal tips for effective Claude Code usage. > ROADMAP. md 概述项目全局，Tasks 文件夹原子化分解步骤，让 Claude Code 通过 One-Shotting 一次性完成 5-8 步任务。> 参考 CrewAI & ChatPRD & TASKMASTER.AI & CodeGuide & Agent-OS，快速发展至 Context Engineering & Agentic Engineering > Harness Engineering & Loop Engineering。

0710 Dev jobs are about to get a hard reset and nobody's ready & 0711 @realcoreychiu: 更新一版我的最新 Vibe Coding 最佳实践方案（性价比版） & @DgudbF28831: 我对用 CLI 跳出 IDE 框架这一步在形式上的探索感到兴奋，是一个迈向更加纯粹的交互的起步，这背后让我真正兴奋的是：通过 Terminal 调用 LLM，我们正在回归一种更本质的交互循环。它不是关于命令行本身，而是关于它所实现的那个纯粹的“意图-执行-审查（Intent-Action-Review）”循环，就是我发现从 IDE 软件跳出来以后，你不用再关心我这个 Code 是怎么写的，你只用关心结果合适不合适，甚至你可以让他去刻意的调用 OS 层的指令，加上 LLM 的能力，轻松当你的贾维斯。

1030 jason liu@jxnlco: Claude Code rewards tinkerers. Background processes that self-heal, custom CLI tools, vision capabilities, and sub-agents that keep your main thread focused on code. & kartikey singh@askwhykartik: Just spent 3 weeks learning Claude Code inside-out.

1228 Software development became boring with Claude Code - 讨论中浮现出一个核心分野：产品导向型开发者 vs 代码纯粹主义者。如果你热爱的是构建产品、解决问题、将想法变为现实，AI 是天赐之物。如果你热爱的是代码本身的智力挑战、那种亲手攻克难题的成就感，确实会感到失落。但也有人提出了更深层的思考："有趣的工作依然存在，只是上移了。决定构建什么、为什么构建、如何融入真实工作流程——这些才是真正的创造。瓶颈从'我能不能实现'变成了'我是否足够理解问题'。" & 20260103 Google Engineer Says Claude Code Rebuilt their System In An Hour - Claude 复现的是代码，不是那一年的认知积累 & 随着 Agent 的时间跨度越来越长，我们都需要学习如何进行长周期规划。> 当 AI 能快速复现已知方案时，人类工程师的核心价值在哪里？答案可能是：在那些需要在不确定性中做出判断、在利益相关者之间达成共识、在模糊需求中提炼本质的工作上。……也许问题不在于 AI 有多强，而在于组织有多慢。

Do a git diff and pretend you're a senior dev doing a code review and you HATE this implementation. What would you criticize? What edge cases am I missing?

20260505 Anthropic's Boris Cherny: Why Coding Is Solved, and What Comes Next - Boris Cherny：Claude Code 之后，写代码正在变成“管理 Agent”

CC Switch - 统一管理你的 AI 编程工具工作流 > 2026 年最主流的 Agent MaaS 网关；0617 Codex 的目标，是让你无论接哪家的模型，开发这件事尽量都在 Codex 里完成。CC Switch 的目标，是让你无论用哪几个开发工具，模型和网络这一层都由它来集中管理。

• farion1231 / cc-switch: A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Gemini CLI & Hermes Agent. Only official website: ccswitch.io

20250715- 调试 Windows 版（Node 环境，注意 Proxifier 的自指规则）：
0）使用 ANTHROPIC_API_KEY 以及其他 MaaS，参考 zebbern / claude-code-guide & Cranot / claude-code-guide（充分利用 AI-Coding 的反身性）：/init & /memory 构建/编辑 CLAUDE.md（大小写敏感），另注意用户目录/项目目录/项目子目录中的 .claude（settings(.local).json 等）& .claude.json & .mcp.json，后使用不同的 settings.xxx.json 配置启动；与 Windows 结合较好（通过 Git Bash 执行）， Agent 执行后根据输出调整行动策略，而在其认为需与人确认的问询交互节点中，选择项之外都可输入其他的提示调整决策路径；

claude --dangerously-skip-permissions（创建 alias 设置其为默认，参考 .bashrc & .bash_profile 文件，需使用 Git Bash 运行）；/doctor & /compact & /clear & /status & /statusline & /resume & /context；Plan Mode（alt-m 两次进入，对应 Todos）；@上下文 & 聊天记录（双击 Esc） & # 添加记忆；拷贝 mcp.json 配置调用 MCP（部分 MCP 报错 -32000: Connection closed）；/install-github-app 启用 Github Actions；
手工配置 autoUpdates 为 true & autoUpdatesProtectedForNative 为 false 无效，后取消配置 CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC 为 1（为启动提速），恢复自动更新；另配置 autoCompactEnabled 为 false 不自动压缩；
 
1.0）使用 ANTHROPIC_API_KEY 连接（不同于 Pro 等计划，无需等待恢复）：Hack 掉的 ANTHROPIC_AUTH_TOKEN 等加后缀 GLM 等暂时关闭；~/.claude/settings.json 切回强制使用如 claude-sonnet-4-5-20250929 & claude-opus-4-1-20250805 & claude-3-5-haiku-20241022；> 1009 为用完已有预付费充值，拟继续调试 SuperClaude-Org / SuperClaude_Framework V4 & davila7 / claude-code-templates，后修复 TONGYI Lingma & Qwen-Coder - QWen Code & Qoder (CLI)/QoderWork & iFlow CLI，采用 Claude Sonnet 4.5 重新构建及审查修改，效果极佳，很快耗尽充值；
赶在预付费充值到期之前，使用 API Key 模式奢侈了一把无需等待恢复的 Claude Code 的 claude-sonnet-4-5-20250929，把上周 Qoder 的 Quest 模式烂尾的两个不大不小的项目调试到位。更加确定，从 AI-Coding 到 AI-Programming，这就是当下在产学研各个方面都在经历的范式迁移。

1.1）配置接入 Kimi-K2（后续 claude-code-router vs 如 fakerybakery / claude-code-kimi-groq 等更薄的 Proxy）：

1.2）配置接入 QWen3-Coder，同上 Hack 方式参考阿里云百炼 & ModelScope（魔搭社区）> 通义千问（Qwen）：ANTHROPIC_BASE_URL=https://dashscope.aliyuncs.com/api/v2/apps/claude-code-proxy & ANTHROPIC_AUTH_TOKEN=sk-YOURKEY & ANTHROPIC MODEL=qwen3-coder-plus；> 相较 qwen-code 采用有每天免费额度的接入配置，该接入配置费用较高（赠送积分很快被消耗）；

1.3）配置接入 GLM-4.5/4.6+：使用 Claude Code 时，MaaS 采用国内大模型，比较 Kimi-K2 vs QWen3-Coder vs GLM-4.5 vs DeepSeek-3.1，哪个更好？ > 在 Claude Code（或类似编码环境）下的 MaaS 应用中，GLM-4.5 被视为整体最佳选择。

1.4）配置接入 DeepSeek-3.1，同上原生支持 Anthropic API，参考深度求索 - DeepSeek & 零一万物 - Yi：另 Anthropic 协议兼容性深度解析：面向Qwen3、Kimi-K2、GLM-4.5 与 Deepseek-V3.1 的开发者指南，DeepSeek 官方接口最佳；
ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic & ANTHROPIC_AUTH_TOKEN=${YOUR_API_KEY} & ANTHROPIC_MODEL=deepseek-chat & ANTHROPIC_SMALL_FAST_MODEL=deepseek-chat；> 使用 think/think hard/ultrathink 需使用 /model 切换至 deepseek-reasoner：不开推理的话模型能力不够强，开推理的话 Token 数比 GLM-4.5 多很多，价格贵太多；

1.5）配置接入 KAT-Coder，同上 Hack 方式参考快手-StreamLake - KAT-Coder & CodeFlicker

1.6）配置接入 MiniMax-M2，同上 Hack 方式参考 Moonshot.AI / 月之暗面 - Kimi - K2/K3 & MiniMax - M2 / Agent & 阶跃星辰 & METASOTA 秘塔科技

1.7）配置接入 Doubao Seed Code，参考豆包 MarsCode - Seed-Coder & TRAE-SOLO/Work & Agent/UI-TARS > DeepFlow

1.8）配置接入 Google Antigravity 的配额，参考 Google - AI Studio & Gemini CLI > AntiGravity & Agents-CLI

2）创建一个 3D 地球浏览的网站（参考 CrewAI & ChatPRD & TASKMASTER.AI & CodeGuide & Agent-OS）：☒ Set up basic HTML structure for 3D Earth website ☒ Implement 3D Earth visualization using Three.js ☒ Add interactive controls for Earth navigation ☒ Add location search functionality ☒ Add satellite imagery overlay ☒ Add markers and POI functionality > 效果较为美观，功能仍有缺陷；Todos 可以异步加入新需求；
2.1）确认安装 sequentialthinking MCP： ☒ Check if sequentialthinking MCP is available ☒ Verify sequentialthinking MCP server is working
2.2）修复 github/context7 mcp 问题，使其正常使用：☒ Diagnose GitHub MCP connection issue ☒ Check GitHub token validity ☒ Test GitHub MCP functionality ☒ Fix configuration loading issue > 调整 .mcp.json；> 后调用 MaaS 报错 API Error: 400 {"error": "type":"invalid_request_error","message":"Invalid request: a max of 128 tools can be supported"}}，配置中保留核心 MCP 服务；
2.3）"think" < "think hard" < "think harder" < "ultrathink" 可以加速 Claude Code 思考；另有 extended-thinking；

0716 VS Code & Cursor 等中安装官方插件配合使用（后成为主要方式）：IDE 中 Terminal 执行，报错 End of central directory record signature not found. Either not a zip file, or file is truncated. > 参考 issues #3598，新版本已修复；> /ide 连接 Cursor 后（多开仅最后一个 connected），同步 Review 上下文，UI 刷新暂不稳定（可直接使用 Terminal）；> 0930 官方插件全面升级：除同命令行在 Node 容器中，Headless mode 会运行 Claude.exe，故同样需配置 Proxifier；插件与命令行复用配置，注意前者可能会缓存 ANTHROPIC_BASE_URL 等配置，导致在不同 MaaS 切换时不稳定；

1031 参考 Obsidian（插件）使用技巧 2023-2025，通过 polyipseity / obsidian-terminal 接入 Claude Code 等 Code Agent CLI；

1128 当下通过国内大模型使用 Claude Code 时，比较 Kimi-K2 系列 vs GLM-4.6 系列 vs MiniMax-M2 系列，以及 vs QWen3-Coder 系列 vs DeepSeek-3.1 系列 vs Doubao Seed Code 系列，哪个最合适？ > 首选 GLM-4.6 系列，候选 MiniMax-M2 系列；> 20260511 订阅 Business 后全面转向 OpenAI Codex CLI/APP & AGENTS.md & Agents SDK / AgentKit，而 Claude Code 采用 GLM-5.1（通过 API，目前国产口碑最好）；另，Claude Code 从 2.1.36 版本开始，会在每个 API 请求的系统提示词开头，加入含随机 5 位十六进制 cch 字段的 x-anthropic-billing-header 内容。由于第三方 API 将其视为普通提示词计算缓存 Key，cch 每次变化会导致缓存命中率归零，引发 Token 消耗暴涨、推理变慢。

202507 调试 NomenAK / SuperClaude V3 > SuperClaude-Org / SuperClaude_Framework V4 - Commands & AI Agents & Behavioral Modes & MCP Servers > Not Software - It's a Framework & Systematic - Not Ad-hoc & Progressive - Not Complex；详见 SuperClaude Framework & SuperPowers；

0807/1106 调试 davila7 / claude-code-templates v1.27.0 - Docker Sandox Provider
0）通过 plugin 的 markpalce 添加 claude-code-templates，先行安装调试 Plugins - Ai Ml Toolkit & Devops Automation：

• Ai Ml Toolkit - AI and Machine Learning development suite with data engineering and model deployment tools > Agent: ai-ml-toolkit:ai-engineer & ai-ml-toolkit:ml-engineer & ai-ml-toolkit:nlp-engineer & ai-ml-toolkit:computer-vision-engineer & ai-ml-toolkit:mlops-engineer
• Devops Automation - DevOps automation suite with CI/CD, infrastructure management, and deployment orchestration > 5 Comands & 4 Agents: devops-automation:cloud-architect & Github Integraion / Docker Mcp
• Project Management Suite - Project management toolkit with sprint planning, task automation, and team collaboration tools > 6 Commands & 3 Agents & Notion Integration / Linear Integration MCP
• Nextjs Vercel Pro - Complete Next.js and Vercel development toolkit with deployment automation and performance optimization > 7 Commands & 2 Agents & Vercel Mcp

1）提示“Deploy model serving infrastructure”调试 Ai Ml Toolkit - ai-ml-toolkit:mlops-engineer（项目目录 ml-serving-infrastructure）：构建 TODO.md：Phase 1: Core Infrastructure > Phase 2: API Endpoints > Phase 3: Monitoring & Observability > Phase 4: A/B Testing Framework > Phase 5: Infrastructure & Deployment > Phase 6: Advanced Features > Phase 7: CI/CD Pipeline > Phase 8: Documentation；技术栈（Technologies Used）：FastAPI & MLflow & Prometheus & Redis & Docker & Kubernetes & Pytest & JWT & Pydantic；完成文件构建后，启动 deploy.sh 部署时暂停止；

2）安装 Devops Automation，报错 Plugin manifest references non-existent files (devops-automation & testing-suite #124，暂卸载；

20250830 axtrur@axtrur: Claude Code 生态项目（Claude Code SDK 成为重要的 Agents 架构）

1103 调试 carlvellotti / claude-code-pm-course > Claude Code for Product Managers
1）在 此基础上，务必将 blocks 翻译后的中文附加在后面

1107 调试 diet103 / claude-code-infrastructure-showcase > Claude Code is a Beast – Tips from 6 Months of Hardcore Use

参考 Specification-Driven Development(SDD) & Spec-Coding，后续如 coleam00 / context-engineering-intro 以及 github / spec-kit 等初期调试尽量使用 Google - AI Studio & Gemini CLI > AntiGravity & Agents-CLI & Github Next & Github Copilot - CLI / APP & Neovim & Aider & Crush & warp & tmux 等有充裕额度的 CLI；

20250717 调试 coleam00 / context-engineering-intro 落地实践 songjiayang / image-rag
1）Context Engineering 模板结构：CLAUDE.md & INITIAL.md & INITIAL_EXAMPLE.md & example 范例；PRPs(Product Requirements Prompts) 目录包括 templates - prp_base.md 基础模板 & EXAMPLE_multi_agent_prp.md 等详细的 PRP；.claude 目录包括 commands - generate-prp.md & execute-prp.md，以及 setting.local.json；> 执行 /generate-prp INITIAL.md 生成 PRPs > /execute-prp PRPs/your-feature-name.md 执行 PRPs 实现（生成项目功能 & 测试代码并修复 Bug）；

0720 拟结合 Cursor.so > Cursor.sh: The AI-first Code Editor & Copilot++ & Cursor SDK & Claude Dev > Cline / Roo Code & Augment Code & Kiro & Kilo & Tessl Cofounder，推进完成 2024科研 心理咨询 with ML-DL & LLM 研究框架 & LIMS - 化学实验室数据自动化采集及管理系统（唐志松 - 常州瑞泉）；> AI-Coding 工具链（Cursor/Kiro+Claude Code）构建 AI-Science/Malleable GIS 内容生成管道；

1031 拟结合工业互联网技术专业 - 工业大数据 & 工业智能 课程建设 - 工业大数据与工业智能（2025-2026第一学期）教学笔记，参考 Claude Code 的反身性课程教学，推进基于 Code Agent CLI & Agent Skills 架构的《工业数据分析与挖掘》新型 AI 课程建设，前期参考 ANTHROPC - Claude Code - Plugins & Agent Skills & Agent SDK；