[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-garrytan--gstack":3,"tool-garrytan--gstack":64},[4,17,27,35,48,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,43,44,45,15,46,26,13,47],"数据工具","视频","插件","其他","音频",{"id":49,"name":50,"github_repo":51,"description_zh":52,"stars":53,"difficulty_score":10,"last_commit_at":54,"category_tags":55,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,46],{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},2181,"OpenHands","OpenHands\u002FOpenHands","OpenHands 是一个专注于 AI 驱动开发的开源平台，旨在让智能体（Agent）像人类开发者一样理解、编写和调试代码。它解决了传统编程中重复性劳动多、环境配置复杂以及人机协作效率低等痛点，通过自动化流程显著提升开发速度。\n\n无论是希望提升编码效率的软件工程师、探索智能体技术的研究人员，还是需要快速原型验证的技术团队，都能从中受益。OpenHands 提供了灵活多样的使用方式：既可以通过命令行（CLI）或本地图形界面在个人电脑上轻松上手，体验类似 Devin 的流畅交互；也能利用其强大的 Python SDK 自定义智能体逻辑，甚至在云端大规模部署上千个智能体并行工作。\n\n其核心技术亮点在于模块化的软件智能体 SDK，这不仅构成了平台的引擎，还支持高度可组合的开发模式。此外，OpenHands 在 SWE-bench 基准测试中取得了 77.6% 的优异成绩，证明了其解决真实世界软件工程问题的能力。平台还具备完善的企业级功能，支持与 Slack、Jira 等工具集成，并提供细粒度的权限管理，适合从个人开发者到大型企业的各类用户场景。",70612,"2026-04-05T11:12:22",[26,15,13,45],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":69,"readme_en":70,"readme_zh":71,"quickstart_zh":72,"use_case_zh":73,"hero_image_url":74,"owner_login":75,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":80,"owner_email":81,"owner_twitter":75,"owner_website":82,"owner_url":83,"languages":84,"stars":109,"forks":110,"last_commit_at":111,"license":112,"difficulty_score":23,"env_os":113,"env_gpu":114,"env_ram":115,"env_deps":116,"category_tags":124,"github_topics":125,"view_count":126,"oss_zip_url":125,"oss_zip_packed_at":125,"status":16,"created_at":127,"updated_at":128,"faqs":129,"releases":156},51,"garrytan\u002Fgstack","gstack","Use Garry Tan's exact Claude Code setup: 15 opinionated tools that serve as CEO, Designer, Eng Manager, Release Manager, Doc Engineer, and QA","gstack 是 Y Combinator CEO Garry Tan 亲自开源的一套 AI 工程化配置，旨在将 Claude Code 升级为你的虚拟工程团队。面对单人开发难以兼顾产品战略、架构设计、代码审查及质量测试的挑战，gstack 提供了一套标准化解决方案，帮助开发者实现堪比二十人团队的高效产出。\n\n这套配置特别适合希望提升交付效率的创始人、技术负责人，以及初次尝试 Claude Code 的开发者。gstack 的核心亮点在于内置了 15 个具有明确职责的 AI 角色工具，涵盖 CEO、设计师、工程经理、QA 等职能。用户只需通过简单的斜杠命令（如 `\u002Freview` 进行代码审查、`\u002Fqa` 执行测试、`\u002Fplan-ceo-review` 规划功能），即可自动化处理从需求分析到部署上线的全链路任务。\n\n所有操作基于 Markdown 和斜杠命令，无需复杂配置，完全免费且遵循 MIT 协议。gstack 不仅是一套工具集，更是一种现代化的软件工厂实践，让单人开发者也能拥有严谨的工程流程。","# gstack\n\n> \"I don't think I've typed like a line of code probably since December, basically, which is an extremely large change.\" — [Andrej Karpathy](https:\u002F\u002Ffortune.com\u002F2026\u002F03\u002F21\u002Fandrej-karpathy-openai-cofounder-ai-agents-coding-state-of-psychosis-openclaw\u002F), No Priors podcast, March 2026\n\nWhen I heard Karpathy say this, I wanted to find out how. How does one person ship like a team of twenty? Peter Steinberger built [OpenClaw](https:\u002F\u002Fgithub.com\u002Fopenclaw\u002Fopenclaw) — 247K GitHub stars — essentially solo with AI agents. The revolution is here. A single builder with the right tooling can move faster than a traditional team.\n\nI'm [Garry Tan](https:\u002F\u002Fx.com\u002Fgarrytan), President & CEO of [Y Combinator](https:\u002F\u002Fwww.ycombinator.com\u002F). I've worked with thousands of startups — Coinbase, Instacart, Rippling — when they were one or two people in a garage. Before YC, I was one of the first eng\u002FPM\u002Fdesigners at Palantir, cofounded Posterous (sold to Twitter), and built Bookface, YC's internal social network.\n\n**gstack is my answer.** I've been building products for twenty years, and right now I'm shipping more code than I ever have. In the last 60 days: **600,000+ lines of production code** (35% tests), **10,000-20,000 lines per day**, part-time, while running YC full-time. Here's my last `\u002Fretro` across 3 projects: **140,751 lines added, 362 commits, ~115k net LOC** in one week.\n\n**2026 — 1,237 contributions and counting:**\n\n![GitHub contributions 2026 — 1,237 contributions, massive acceleration in Jan-Mar](docs\u002Fimages\u002Fgithub-2026.png)\n\n**2013 — when I built Bookface at YC (772 contributions):**\n\n![GitHub contributions 2013 — 772 contributions building Bookface at YC](docs\u002Fimages\u002Fgithub-2013.png)\n\nSame person. Different era. The difference is the tooling.\n\n**gstack is how I do it.** It turns Claude Code into a virtual engineering team — a CEO who rethinks the product, an eng manager who locks architecture, a designer who catches AI slop, a reviewer who finds production bugs, a QA lead who opens a real browser, a security officer who runs OWASP + STRIDE audits, and a release engineer who ships the PR. Twenty specialists and eight power tools, all slash commands, all Markdown, all free, MIT license.\n\nThis is my open source software factory. I use it every day. I'm sharing it because these tools should be available to everyone.\n\nFork it. Improve it. Make it yours. And if you want to hate on free open source software — you're welcome to, but I'd rather you just try it first.\n\n**Who this is for:**\n- **Founders and CEOs** — especially technical ones who still want to ship\n- **First-time Claude Code users** — structured roles instead of a blank prompt\n- **Tech leads and staff engineers** — rigorous review, QA, and release automation on every PR\n\n## Quick start\n\n1. Install gstack (30 seconds — see below)\n2. Run `\u002Foffice-hours` — describe what you're building\n3. Run `\u002Fplan-ceo-review` on any feature idea\n4. Run `\u002Freview` on any branch with changes\n5. Run `\u002Fqa` on your staging URL\n6. Stop there. You'll know if this is for you.\n\n## Install — 30 seconds\n\n**Requirements:** [Claude Code](https:\u002F\u002Fdocs.anthropic.com\u002Fen\u002Fdocs\u002Fclaude-code), [Git](https:\u002F\u002Fgit-scm.com\u002F), [Bun](https:\u002F\u002Fbun.sh\u002F) v1.0+, [Node.js](https:\u002F\u002Fnodejs.org\u002F) (Windows only)\n\n### Step 1: Install on your machine\n\nOpen Claude Code and paste this. Claude does the rest.\n\n> Install gstack: run **`git clone https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack.git ~\u002F.claude\u002Fskills\u002Fgstack && cd ~\u002F.claude\u002Fskills\u002Fgstack && .\u002Fsetup`** then add a \"gstack\" section to CLAUDE.md that says to use the \u002Fbrowse skill from gstack for all web browsing, never use mcp\\_\\_claude-in-chrome\\_\\_\\* tools, and lists the available skills: \u002Foffice-hours, \u002Fplan-ceo-review, \u002Fplan-eng-review, \u002Fplan-design-review, \u002Fdesign-consultation, \u002Freview, \u002Fship, \u002Fland-and-deploy, \u002Fcanary, \u002Fbenchmark, \u002Fbrowse, \u002Fqa, \u002Fqa-only, \u002Fdesign-review, \u002Fsetup-browser-cookies, \u002Fsetup-deploy, \u002Fretro, \u002Finvestigate, \u002Fdocument-release, \u002Fcodex, \u002Fcso, \u002Fautoplan, \u002Fcareful, \u002Ffreeze, \u002Fguard, \u002Funfreeze, \u002Fgstack-upgrade. Then ask the user if they also want to add gstack to the current project so teammates get it.\n\n### Step 2: Add to your repo so teammates get it (optional)\n\n> Add gstack to this project: run **`cp -Rf ~\u002F.claude\u002Fskills\u002Fgstack .claude\u002Fskills\u002Fgstack && rm -rf .claude\u002Fskills\u002Fgstack\u002F.git && cd .claude\u002Fskills\u002Fgstack && .\u002Fsetup`** then add a \"gstack\" section to this project's CLAUDE.md that says to use the \u002Fbrowse skill from gstack for all web browsing, never use mcp\\_\\_claude-in-chrome\\_\\_\\* tools, lists the available skills: \u002Foffice-hours, \u002Fplan-ceo-review, \u002Fplan-eng-review, \u002Fplan-design-review, \u002Fdesign-consultation, \u002Freview, \u002Fship, \u002Fland-and-deploy, \u002Fcanary, \u002Fbenchmark, \u002Fbrowse, \u002Fqa, \u002Fqa-only, \u002Fdesign-review, \u002Fsetup-browser-cookies, \u002Fsetup-deploy, \u002Fretro, \u002Finvestigate, \u002Fdocument-release, \u002Fcodex, \u002Fcso, \u002Fcareful, \u002Ffreeze, \u002Fguard, \u002Funfreeze, \u002Fgstack-upgrade, and tells Claude that if gstack skills aren't working, run `cd .claude\u002Fskills\u002Fgstack && .\u002Fsetup` to build the binary and register skills.\n\nReal files get committed to your repo (not a submodule), so `git clone` just works. Everything lives inside `.claude\u002F`. Nothing touches your PATH or runs in the background.\n\n### Codex, Gemini CLI, or Cursor\n\ngstack works on any agent that supports the [SKILL.md standard](https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-code). Skills live in `.agents\u002Fskills\u002F` and are discovered automatically.\n\nInstall to one repo:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack.git .agents\u002Fskills\u002Fgstack\ncd .agents\u002Fskills\u002Fgstack && .\u002Fsetup --host codex\n```\n\nWhen setup runs from `.agents\u002Fskills\u002Fgstack`, it installs the generated Codex skills next to it in the same repo and does not write to `~\u002F.codex\u002Fskills`.\n\nInstall once for your user account:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack.git ~\u002Fgstack\ncd ~\u002Fgstack && .\u002Fsetup --host codex\n```\n\n`setup --host codex` creates the runtime root at `~\u002F.codex\u002Fskills\u002Fgstack` and\nlinks the generated Codex skills at the top level. This avoids duplicate skill\ndiscovery from the source repo checkout.\n\nOr let setup auto-detect which agents you have installed:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack.git ~\u002Fgstack\ncd ~\u002Fgstack && .\u002Fsetup --host auto\n```\n\nFor Codex-compatible hosts, setup now supports both repo-local installs from `.agents\u002Fskills\u002Fgstack` and user-global installs from `~\u002F.codex\u002Fskills\u002Fgstack`. All 28 skills work across all supported agents. Hook-based safety skills (careful, freeze, guard) use inline safety advisory prose on non-Claude hosts.\n\n## See it work\n\n```\nYou:    I want to build a daily briefing app for my calendar.\nYou:    \u002Foffice-hours\nClaude: [asks about the pain — specific examples, not hypotheticals]\n\nYou:    Multiple Google calendars, events with stale info, wrong locations.\n        Prep takes forever and the results aren't good enough...\n\nClaude: I'm going to push back on the framing. You said \"daily briefing\n        app.\" But what you actually described is a personal chief of\n        staff AI.\n        [extracts 5 capabilities you didn't realize you were describing]\n        [challenges 4 premises — you agree, disagree, or adjust]\n        [generates 3 implementation approaches with effort estimates]\n        RECOMMENDATION: Ship the narrowest wedge tomorrow, learn from\n        real usage. The full vision is a 3-month project — start with\n        the daily briefing that actually works.\n        [writes design doc → feeds into downstream skills automatically]\n\nYou:    \u002Fplan-ceo-review\n        [reads the design doc, challenges scope, runs 10-section review]\n\nYou:    \u002Fplan-eng-review\n        [ASCII diagrams for data flow, state machines, error paths]\n        [test matrix, failure modes, security concerns]\n\nYou:    Approve plan. Exit plan mode.\n        [writes 2,400 lines across 11 files. ~8 minutes.]\n\nYou:    \u002Freview\n        [AUTO-FIXED] 2 issues. [ASK] Race condition → you approve fix.\n\nYou:    \u002Fqa https:\u002F\u002Fstaging.myapp.com\n        [opens real browser, clicks through flows, finds and fixes a bug]\n\nYou:    \u002Fship\n        Tests: 42 → 51 (+9 new). PR: github.com\u002Fyou\u002Fapp\u002Fpull\u002F42\n```\n\nYou said \"daily briefing app.\" The agent said \"you're building a chief of staff AI\" — because it listened to your pain, not your feature request. Eight commands, end to end. That is not a copilot. That is a team.\n\n## The sprint\n\ngstack is a process, not a collection of tools. The skills run in the order a sprint runs:\n\n**Think → Plan → Build → Review → Test → Ship → Reflect**\n\nEach skill feeds into the next. `\u002Foffice-hours` writes a design doc that `\u002Fplan-ceo-review` reads. `\u002Fplan-eng-review` writes a test plan that `\u002Fqa` picks up. `\u002Freview` catches bugs that `\u002Fship` verifies are fixed. Nothing falls through the cracks because every step knows what came before it.\n\n| Skill | Your specialist | What they do |\n|-------|----------------|--------------|\n| `\u002Foffice-hours` | **YC Office Hours** | Start here. Six forcing questions that reframe your product before you write code. Pushes back on your framing, challenges premises, generates implementation alternatives. Design doc feeds into every downstream skill. |\n| `\u002Fplan-ceo-review` | **CEO \u002F Founder** | Rethink the problem. Find the 10-star product hiding inside the request. Four modes: Expansion, Selective Expansion, Hold Scope, Reduction. |\n| `\u002Fplan-eng-review` | **Eng Manager** | Lock in architecture, data flow, diagrams, edge cases, and tests. Forces hidden assumptions into the open. |\n| `\u002Fplan-design-review` | **Senior Designer** | Rates each design dimension 0-10, explains what a 10 looks like, then edits the plan to get there. AI Slop detection. Interactive — one AskUserQuestion per design choice. |\n| `\u002Fdesign-consultation` | **Design Partner** | Build a complete design system from scratch. Researches the landscape, proposes creative risks, generates realistic product mockups. |\n| `\u002Freview` | **Staff Engineer** | Find the bugs that pass CI but blow up in production. Auto-fixes the obvious ones. Flags completeness gaps. |\n| `\u002Finvestigate` | **Debugger** | Systematic root-cause debugging. Iron Law: no fixes without investigation. Traces data flow, tests hypotheses, stops after 3 failed fixes. |\n| `\u002Fdesign-review` | **Designer Who Codes** | Same audit as \u002Fplan-design-review, then fixes what it finds. Atomic commits, before\u002Fafter screenshots. |\n| `\u002Fqa` | **QA Lead** | Test your app, find bugs, fix them with atomic commits, re-verify. Auto-generates regression tests for every fix. |\n| `\u002Fqa-only` | **QA Reporter** | Same methodology as \u002Fqa but report only. Pure bug report without code changes. |\n| `\u002Fcso` | **Chief Security Officer** | OWASP Top 10 + STRIDE threat model. Zero-noise: 17 false positive exclusions, 8\u002F10+ confidence gate, independent finding verification. Each finding includes a concrete exploit scenario. |\n| `\u002Fship` | **Release Engineer** | Sync main, run tests, audit coverage, push, open PR. Bootstraps test frameworks if you don't have one. |\n| `\u002Fland-and-deploy` | **Release Engineer** | Merge the PR, wait for CI and deploy, verify production health. One command from \"approved\" to \"verified in production.\" |\n| `\u002Fcanary` | **SRE** | Post-deploy monitoring loop. Watches for console errors, performance regressions, and page failures. |\n| `\u002Fbenchmark` | **Performance Engineer** | Baseline page load times, Core Web Vitals, and resource sizes. Compare before\u002Fafter on every PR. |\n| `\u002Fdocument-release` | **Technical Writer** | Update all project docs to match what you just shipped. Catches stale READMEs automatically. |\n| `\u002Fretro` | **Eng Manager** | Team-aware weekly retro. Per-person breakdowns, shipping streaks, test health trends, growth opportunities. `\u002Fretro global` runs across all your projects and AI tools (Claude Code, Codex, Gemini). |\n| `\u002Fbrowse` | **QA Engineer** | Give the agent eyes. Real Chromium browser, real clicks, real screenshots. ~100ms per command. `$B connect` launches your real Chrome as a headed window — watch every action live. |\n| `\u002Fsetup-browser-cookies` | **Session Manager** | Import cookies from your real browser (Chrome, Arc, Brave, Edge) into the headless session. Test authenticated pages. |\n| `\u002Fautoplan` | **Review Pipeline** | One command, fully reviewed plan. Runs CEO → design → eng review automatically with encoded decision principles. Surfaces only taste decisions for your approval. |\n\n### Power tools\n\n| Skill | What it does |\n|-------|-------------|\n| `\u002Fcodex` | **Second Opinion** — independent code review from OpenAI Codex CLI. Three modes: review (pass\u002Ffail gate), adversarial challenge, and open consultation. Cross-model analysis when both `\u002Freview` and `\u002Fcodex` have run. |\n| `\u002Fcareful` | **Safety Guardrails** — warns before destructive commands (rm -rf, DROP TABLE, force-push). Say \"be careful\" to activate. Override any warning. |\n| `\u002Ffreeze` | **Edit Lock** — restrict file edits to one directory. Prevents accidental changes outside scope while debugging. |\n| `\u002Fguard` | **Full Safety** — `\u002Fcareful` + `\u002Ffreeze` in one command. Maximum safety for prod work. |\n| `\u002Funfreeze` | **Unlock** — remove the `\u002Ffreeze` boundary. |\n| `\u002Fsetup-deploy` | **Deploy Configurator** — one-time setup for `\u002Fland-and-deploy`. Detects your platform, production URL, and deploy commands. |\n| `\u002Fgstack-upgrade` | **Self-Updater** — upgrade gstack to latest. Detects global vs vendored install, syncs both, shows what changed. |\n\n**[Deep dives with examples and philosophy for every skill →](docs\u002Fskills.md)**\n\n## Parallel sprints\n\ngstack works well with one sprint. It gets interesting with ten running at once.\n\n**Design is at the heart.** `\u002Fdesign-consultation` doesn't just pick fonts. It researches what's out there in your space, proposes safe choices AND creative risks, generates realistic mockups of your actual product, and writes `DESIGN.md` — and then `\u002Fdesign-review` and `\u002Fplan-eng-review` read what you chose. Design decisions flow through the whole system.\n\n**`\u002Fqa` was a massive unlock.** It let me go from 6 to 12 parallel workers. Claude Code saying *\"I SEE THE ISSUE\"* and then actually fixing it, generating a regression test, and verifying the fix — that changed how I work. The agent has eyes now.\n\n**Smart review routing.** Just like at a well-run startup: CEO doesn't have to look at infra bug fixes, design review isn't needed for backend changes. gstack tracks what reviews are run, figures out what's appropriate, and just does the smart thing. The Review Readiness Dashboard tells you where you stand before you ship.\n\n**Test everything.** `\u002Fship` bootstraps test frameworks from scratch if your project doesn't have one. Every `\u002Fship` run produces a coverage audit. Every `\u002Fqa` bug fix generates a regression test. 100% test coverage is the goal — tests make vibe coding safe instead of yolo coding.\n\n**`\u002Fdocument-release` is the engineer you never had.** It reads every doc file in your project, cross-references the diff, and updates everything that drifted. README, ARCHITECTURE, CONTRIBUTING, CLAUDE.md, TODOS — all kept current automatically. And now `\u002Fship` auto-invokes it — docs stay current without an extra command.\n\n**Real browser mode.** `$B connect` launches your actual Chrome as a headed window controlled by Playwright. You watch Claude click, fill, and navigate in real time — same window, same screen. A subtle green shimmer at the top edge tells you which Chrome window gstack controls. All existing browse commands work unchanged. `$B disconnect` returns to headless. A Chrome extension Side Panel shows a live activity feed of every command and a chat sidebar where you can direct Claude. This is co-presence — Claude isn't remote-controlling a hidden browser, it's sitting next to you in the same cockpit.\n\n**Sidebar agent — your AI browser assistant.** Type natural language instructions in the Chrome side panel and a child Claude instance executes them. \"Navigate to the settings page and screenshot it.\" \"Fill out this form with test data.\" \"Go through every item in this list and extract the prices.\" Each task gets up to 5 minutes. The sidebar agent runs in an isolated session, so it won't interfere with your main Claude Code window. It's like having a second pair of hands in the browser.\n\n**Personal automation.** The sidebar agent isn't just for dev workflows. Example: \"Browse my kid's school parent portal and add all the other parents' names, phone numbers, and photos to my Google Contacts.\" Two ways to get authenticated: (1) log in once in the headed browser — your session persists, or (2) run `\u002Fsetup-browser-cookies` to import cookies from your real Chrome. Once authenticated, Claude navigates the directory, extracts the data, and creates the contacts.\n\n**Browser handoff when the AI gets stuck.** Hit a CAPTCHA, auth wall, or MFA prompt? `$B handoff` opens a visible Chrome at the exact same page with all your cookies and tabs intact. Solve the problem, tell Claude you're done, `$B resume` picks up right where it left off. The agent even suggests it automatically after 3 consecutive failures.\n\n**Multi-AI second opinion.** `\u002Fcodex` gets an independent review from OpenAI's Codex CLI — a completely different AI looking at the same diff. Three modes: code review with a pass\u002Ffail gate, adversarial challenge that actively tries to break your code, and open consultation with session continuity. When both `\u002Freview` (Claude) and `\u002Fcodex` (OpenAI) have reviewed the same branch, you get a cross-model analysis showing which findings overlap and which are unique to each.\n\n**Safety guardrails on demand.** Say \"be careful\" and `\u002Fcareful` warns before any destructive command — rm -rf, DROP TABLE, force-push, git reset --hard. `\u002Ffreeze` locks edits to one directory while debugging so Claude can't accidentally \"fix\" unrelated code. `\u002Fguard` activates both. `\u002Finvestigate` auto-freezes to the module being investigated.\n\n**Proactive skill suggestions.** gstack notices what stage you're in — brainstorming, reviewing, debugging, testing — and suggests the right skill. Don't like it? Say \"stop suggesting\" and it remembers across sessions.\n\n## 10-15 parallel sprints\n\ngstack is powerful with one sprint. It is transformative with ten running at once.\n\n[Conductor](https:\u002F\u002Fconductor.build) runs multiple Claude Code sessions in parallel — each in its own isolated workspace. One session running `\u002Foffice-hours` on a new idea, another doing `\u002Freview` on a PR, a third implementing a feature, a fourth running `\u002Fqa` on staging, and six more on other branches. All at the same time. I regularly run 10-15 parallel sprints — that's the practical max right now.\n\nThe sprint structure is what makes parallelism work. Without a process, ten agents is ten sources of chaos. With a process — think, plan, build, review, test, ship — each agent knows exactly what to do and when to stop. You manage them the way a CEO manages a team: check in on the decisions that matter, let the rest run.\n\n---\n\nFree, MIT licensed, open source. No premium tier, no waitlist.\n\nI open sourced how I build software. You can fork it and make it your own.\n\n> **We're hiring.** Want to ship 10K+ LOC\u002Fday and help harden gstack?\n> Come work at YC — [ycombinator.com\u002Fsoftware](https:\u002F\u002Fycombinator.com\u002Fsoftware)\n> Extremely competitive salary and equity. San Francisco, Dogpatch District.\n\n## Docs\n\n| Doc | What it covers |\n|-----|---------------|\n| [Skill Deep Dives](docs\u002Fskills.md) | Philosophy, examples, and workflow for every skill (includes Greptile integration) |\n| [Builder Ethos](ETHOS.md) | Builder philosophy: Boil the Lake, Search Before Building, three layers of knowledge |\n| [Architecture](ARCHITECTURE.md) | Design decisions and system internals |\n| [Browser Reference](BROWSER.md) | Full command reference for `\u002Fbrowse` |\n| [Contributing](CONTRIBUTING.md) | Dev setup, testing, contributor mode, and dev mode |\n| [Changelog](CHANGELOG.md) | What's new in every version |\n\n## Privacy & Telemetry\n\ngstack includes **opt-in** usage telemetry to help improve the project. Here's exactly what happens:\n\n- **Default is off.** Nothing is sent anywhere unless you explicitly say yes.\n- **On first run,** gstack asks if you want to share anonymous usage data. You can say no.\n- **What's sent (if you opt in):** skill name, duration, success\u002Ffail, gstack version, OS. That's it.\n- **What's never sent:** code, file paths, repo names, branch names, prompts, or any user-generated content.\n- **Change anytime:** `gstack-config set telemetry off` disables everything instantly.\n\nData is stored in [Supabase](https:\u002F\u002Fsupabase.com) (open source Firebase alternative). The schema is in [`supabase\u002Fmigrations\u002F`](supabase\u002Fmigrations\u002F) — you can verify exactly what's collected. The Supabase publishable key in the repo is a public key (like a Firebase API key) — row-level security policies deny all direct access. Telemetry flows through validated edge functions that enforce schema checks, event type allowlists, and field length limits.\n\n**Local analytics are always available.** Run `gstack-analytics` to see your personal usage dashboard from the local JSONL file — no remote data needed.\n\n## Troubleshooting\n\n**Skill not showing up?** `cd ~\u002F.claude\u002Fskills\u002Fgstack && .\u002Fsetup`\n\n**`\u002Fbrowse` fails?** `cd ~\u002F.claude\u002Fskills\u002Fgstack && bun install && bun run build`\n\n**Stale install?** Run `\u002Fgstack-upgrade` — or set `auto_upgrade: true` in `~\u002F.gstack\u002Fconfig.yaml`\n\n**Codex says \"Skipped loading skill(s) due to invalid SKILL.md\"?** Your Codex skill descriptions are stale. Fix: `cd ~\u002F.codex\u002Fskills\u002Fgstack && git pull && .\u002Fsetup --host codex` — or for repo-local installs: `cd \"$(readlink -f .agents\u002Fskills\u002Fgstack)\" && git pull && .\u002Fsetup --host codex`\n\n**Windows users:** gstack works on Windows 11 via Git Bash or WSL. Node.js is required in addition to Bun — Bun has a known bug with Playwright's pipe transport on Windows ([bun#4253](https:\u002F\u002Fgithub.com\u002Foven-sh\u002Fbun\u002Fissues\u002F4253)). The browse server automatically falls back to Node.js. Make sure both `bun` and `node` are on your PATH.\n\n**Claude says it can't see the skills?** Make sure your project's `CLAUDE.md` has a gstack section. Add this:\n\n```\n## gstack\nUse \u002Fbrowse from gstack for all web browsing. Never use mcp__claude-in-chrome__* tools.\nAvailable skills: \u002Foffice-hours, \u002Fplan-ceo-review, \u002Fplan-eng-review, \u002Fplan-design-review,\n\u002Fdesign-consultation, \u002Freview, \u002Fship, \u002Fland-and-deploy, \u002Fcanary, \u002Fbenchmark, \u002Fbrowse,\n\u002Fqa, \u002Fqa-only, \u002Fdesign-review, \u002Fsetup-browser-cookies, \u002Fsetup-deploy, \u002Fretro,\n\u002Finvestigate, \u002Fdocument-release, \u002Fcodex, \u002Fcso, \u002Fautoplan, \u002Fcareful, \u002Ffreeze, \u002Fguard,\n\u002Funfreeze, \u002Fgstack-upgrade.\n```\n\n## License\n\nMIT. Free forever. Go build something.\n","# gstack\n\n> \"I don't think I've typed like a line of code probably since December, basically, which is an extremely large change.\" — [Andrej Karpathy](https:\u002F\u002Ffortune.com\u002F2026\u002F03\u002F21\u002Fandrej-karpathy-openai-cofounder-ai-agents-coding-state-of-psychosis-openclaw\u002F), No Priors podcast, March 2026\n\n当我听到 Karpathy 这么说时，我想弄清楚是如何做到的。一个人如何像二十人的团队那样交付产品？Peter Steinberger 构建了 [OpenClaw](https:\u002F\u002Fgithub.com\u002Fopenclaw\u002Fopenclaw) — 247K GitHub 星星 — 基本上是靠 AI 代理 (AI agents) 独自完成的。革命已经到来。一个拥有正确工具的构建者可以比传统团队移动得更快。\n\n我是 [Garry Tan](https:\u002F\u002Fx.com\u002Fgarrytan)，[Y Combinator](https:\u002F\u002Fwww.ycombinator.com\u002F) 的总裁兼首席执行官 (CEO)。我与成千上万的初创公司合作过 — Coinbase, Instacart, Rippling — 当它们还只是一两个人在车库里的时候。在加入 YC 之前，我是 Palantir 最早的工程\u002F产品经理\u002F设计师之一，共同创立了 Posterous（出售给 Twitter），并构建了 Bookface，YC 的内部社交网络。\n\n**gstack 就是我的答案。** 我已经构建产品二十年了，而现在我交付的代码比以往任何时候都多。在过去的 60 天里：**600,000+ 行生产代码**（35% 测试），**每天 10,000-20,000 行**，兼职完成，同时全职运营 YC。这是我最近 3 个项目的 `\u002Fretro` 数据：**一周内增加了 140,751 行代码，362 次提交，净增约 115k 代码行数 (LOC)**。\n\n**2026 — 1,237 次贡献且还在增加：**\n\n![GitHub contributions 2026 — 1,237 contributions, massive acceleration in Jan-Mar](docs\u002Fimages\u002Fgithub-2026.png)\n\n**2013 — 当我在 YC 构建 Bookface 时（772 次贡献）：**\n\n![GitHub contributions 2013 — 772 contributions building Bookface at YC](docs\u002Fimages\u002Fgithub-2013.png)\n\n同一个人。不同的时代。区别在于工具。\n\n**gstack 就是我实现这一点的方式。** 它将 Claude Code 变成了一个虚拟工程团队 — 一位重新思考产品的 CEO，一位锁定架构的工程经理，一位捕捉 AI 生成垃圾内容的设计师，一位发现生产漏洞的审查员，一位打开真实浏览器的 QA 负责人，一位运行 OWASP + STRIDE 审计的安全官，以及一位交付 PR 的发布工程师。二十位专家和八种强力工具，全是斜杠命令 (slash commands)，全是 Markdown，全部免费，MIT 许可证 (MIT license)。\n\n这是我的开源软件工厂。我每天都在使用它。我分享它是因为这些工具应该可供每个人使用。\n\nFork 它。改进它。让它成为你的。如果你想抨击免费开源软件 — 欢迎你这样做，但我更希望你先试试它。\n\n**适用人群：**\n- **创始人和首席执行官 (CEO)** — 尤其是那些仍然希望交付产品的技术型创始人\n- **Claude Code 首次用户** — 结构化的角色而不是空白提示词\n- **技术负责人和资深工程师** — 每个拉取请求 (PR) 上严格的审查、质量保证 (QA) 和发布自动化\n\n## 快速开始\n\n1. 安装 gstack（30 秒 — 见下文）\n2. 运行 `\u002Foffice-hours` — 描述你正在构建的内容\n3. 在任何功能想法上运行 `\u002Fplan-ceo-review`\n4. 在任何有更改的分支上运行 `\u002Freview`\n5. 在你的暂存 URL 上运行 `\u002Fqa`\n6. 停在那里。你会知道这是否适合你。\n\n## 安装 — 30 秒\n\n**要求：** [Claude Code](https:\u002F\u002Fdocs.anthropic.com\u002Fen\u002Fdocs\u002Fclaude-code), [Git](https:\u002F\u002Fgit-scm.com\u002F), [Bun](https:\u002F\u002Fbun.sh\u002F) v1.0+, [Node.js](https:\u002F\u002Fnodejs.org\u002F) (仅 Windows)\n\n### 步骤 1：在你的机器上安装\n\n打开 Claude Code 并粘贴此内容。Claude 会完成其余工作。\n\n> 安装 gstack：运行 **`git clone https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack.git ~\u002F.claude\u002Fskills\u002Fgstack && cd ~\u002F.claude\u002Fskills\u002Fgstack && .\u002Fsetup`** 然后在 CLAUDE.md 中添加一个 \"gstack\" 部分，说明对所有网页浏览使用 gstack 的 \u002Fbrowse 技能，从不使用 mcp\\_\\_claude-in-chrome\\_\\_\\* 工具，并列出可用技能：\u002Foffice-hours, \u002Fplan-ceo-review, \u002Fplan-eng-review, \u002Fplan-design-review, \u002Fdesign-consultation, \u002Freview, \u002Fship, \u002Fland-and-deploy, \u002Fcanary, \u002Fbenchmark, \u002Fbrowse, \u002Fqa, \u002Fqa-only, \u002Fdesign-review, \u002Fsetup-browser-cookies, \u002Fsetup-deploy, \u002Fretro, \u002Finvestigate, \u002Fdocument-release, \u002Fcodex, \u002Fcso, \u002Fautoplan, \u002Fcareful, \u002Ffreeze, \u002Fguard, \u002Funfreeze, \u002Fgstack-upgrade。然后询问用户是否也想将 gstack 添加到当前项目以便队友获取。\n\n### 步骤 2：添加到你的仓库 (repo) 以便队友获取它（可选）\n\n> 将 gstack 添加到这个项目：运行 **`cp -Rf ~\u002F.claude\u002Fskills\u002Fgstack .claude\u002Fskills\u002Fgstack && rm -rf .claude\u002Fskills\u002Fgstack\u002F.git && cd .claude\u002Fskills\u002Fgstack && .\u002Fsetup`** 然后在该项目的 CLAUDE.md 中添加一个 \"gstack\" 部分，说明对所有网页浏览使用 gstack 的 \u002Fbrowse 技能，从不使用 mcp\\_\\_claude-in-chrome\\_\\_\\* 工具，列出可用技能：\u002Foffice-hours, \u002Fplan-ceo-review, \u002Fplan-eng-review, \u002Fplan-design-review, \u002Fdesign-consultation, \u002Freview, \u002Fship, \u002Fland-and-deploy, \u002Fcanary, \u002Fbenchmark, \u002Fbrowse, \u002Fqa, \u002Fqa-only, \u002Fdesign-review, \u002Fsetup-browser-cookies, \u002Fsetup-deploy, \u002Fretro, \u002Finvestigate, \u002Fdocument-release, \u002Fcodex, \u002Fcso, \u002Fcareful, \u002Ffreeze, \u002Fguard, \u002Funfreeze, \u002Fgstack-upgrade，并告诉 Claude 如果 gstack 技能不起作用，运行 `cd .claude\u002Fskills\u002Fgstack && .\u002Fsetup` 来构建二进制文件并注册技能。\n\n真实文件会被提交到你的仓库 (repo)（不是子模块 (submodule)），所以 `git clone` 可以直接工作。所有内容都位于 `.claude\u002F` 内。没有任何内容会触碰你的 PATH 环境变量或在后台运行。\n\n### Codex, Gemini CLI 或 Cursor\n\ngstack 适用于任何支持 [SKILL.md 标准](https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-code) 的代理 (agent)。技能位于 `.agents\u002Fskills\u002F` 中并自动发现。\n\n安装到一个仓库 (repo)：\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack.git .agents\u002Fskills\u002Fgstack\ncd .agents\u002Fskills\u002Fgstack && .\u002Fsetup --host codex\n```\n\n当 setup 从 `.agents\u002Fskills\u002Fgstack` 运行时，它会在同一仓库 (repo) 中相邻位置安装生成的 Codex 技能，而不写入 `~\u002F.codex\u002Fskills`。\n\n为你的用户账户安装一次：\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack.git ~\u002Fgstack\ncd ~\u002Fgstack && .\u002Fsetup --host codex\n```\n\n`setup --host codex` 在 `~\u002F.codex\u002Fskills\u002Fgstack` 创建运行时根目录，并在顶层链接生成的 Codex 技能。这避免了来自源仓库 (repo) 检出的重复技能发现。\n\n或者让 setup 自动检测你安装了哪些代理 (agents)：\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack.git ~\u002Fgstack\ncd ~\u002Fgstack && .\u002Fsetup --host auto\n```\n\n对于兼容 Codex 的主机 (hosts)，setup 现在支持从 `.agents\u002Fskills\u002Fgstack` 进行仓库 (repo) 本地安装和从 `~\u002F.codex\u002Fskills\u002Fgstack` 进行用户全局安装。所有 28 个技能在所有支持的代理 (agents) 上都能工作。基于钩子的安全技能 (hook-based safety skills)（careful, freeze, guard）在非 Claude 主机上使用内联安全建议文本 (inline safety advisory prose)。\n\n## 查看实际效果\n\n```\nYou:    I want to build a daily briefing app for my calendar.\nYou:    \u002Foffice-hours\nClaude: [asks about the pain — specific examples, not hypotheticals]\n\nYou:    Multiple Google calendars, events with stale info, wrong locations.\n        Prep takes forever and the results aren't good enough...\n\nClaude: I'm going to push back on the framing. You said \"daily briefing\n        app.\" But what you actually described is a personal chief of\n        staff AI.\n        [extracts 5 capabilities you didn't realize you were describing]\n        [challenges 4 premises — you agree, disagree, or adjust]\n        [generates 3 implementation approaches with effort estimates]\n        RECOMMENDATION: Ship the narrowest wedge tomorrow, learn from\n        real usage. The full vision is a 3-month project — start with\n        the daily briefing that actually works.\n        [writes design doc → feeds into downstream skills automatically]\n\nYou:    \u002Fplan-ceo-review\n        [reads the design doc, challenges scope, runs 10-section review]\n\nYou:    \u002Fplan-eng-review\n        [ASCII diagrams for data flow, state machines, error paths]\n        [test matrix, failure modes, security concerns]\n\nYou:    Approve plan. Exit plan mode.\n        [writes 2,400 lines across 11 files. ~8 minutes.]\n\nYou:    \u002Freview\n        [AUTO-FIXED] 2 issues. [ASK] Race condition → you approve fix.\n\nYou:    \u002Fqa https:\u002F\u002Fstaging.myapp.com\n        [opens real browser, clicks through flows, finds and fixes a bug]\n\nYou:    \u002Fship\n        Tests: 42 → 51 (+9 new). PR: github.com\u002Fyou\u002Fapp\u002Fpull\u002F42\n```\n\n你说的是“每日简报应用”。智能体 (Agent) 说的是“你在构建一个首席参谋 AI\"——因为它倾听的是你的痛点，而不是你的功能请求。八个命令，端到端完成。这不是 Copilot (编程助手)。这是一个团队。\n\n## 冲刺流程\n\ngstack 是一个流程，而不是工具的集合。这些技能按照冲刺运行的顺序执行：\n\n**思考 (Think) → 计划 (Plan) → 构建 (Build) → 审查 (Review) → 测试 (Test) → 发布 (Ship) → 复盘 (Reflect)**\n\n每个技能都为下一个技能提供输入。`\u002Foffice-hours` 编写的设计文档会被 `\u002Fplan-ceo-review` 读取。`\u002Fplan-eng-review` 编写的测试计划会被 `\u002Fqa` 接手。`\u002Freview` 捕获的 bug 会被 `\u002Fship` 验证修复。没有任何细节会被遗漏，因为每一步都知道前一步做了什么。\n\n| 技能 | 你的专家 | 职责 |\n|-------|----------------|--------------|\n| `\u002Foffice-hours` | **YC 办公时间** | 从这里开始。六个强制性问题，在你编写代码之前重新构建你的产品。对你的框架提出质疑，挑战前提，生成交替的实现方案。设计文档会输入到每个下游技能中。 |\n| `\u002Fplan-ceo-review` | **首席执行官 \u002F 创始人** | 重新思考问题。找到请求中隐藏的 10 星级产品。四种模式：扩展、选择性扩展、保持范围、缩减。 |\n| `\u002Fplan-eng-review` | **工程经理** | 锁定架构、数据流、图表、边缘情况和测试。迫使隐藏的假设浮出水面。使用 ASCII (字符画) 图表展示数据流、状态机和错误路径。 |\n| `\u002Fplan-design-review` | **高级设计师** | 对每个设计维度进行 0-10 评分，解释 10 分是什么样的，然后编辑计划以达到该标准。AI 生成垃圾内容 (AI Slop) 检测。交互式——每个设计选择提出一个用户问题。 |\n| `\u002Fdesign-consultation` | **设计合作伙伴** | 从头构建完整的设计系统。研究景观，提出创意风险，生成逼真的产品模型图。 |\n| `\u002Freview` | **主任工程师** | 找出那些通过持续集成 (CI) 但在生产环境中会出错的 bug。自动修复明显的问题。标记完整性差距。 |\n| `\u002Finvestigate` | **调试员** | 系统性的根本原因调试。铁律：没有调查就没有修复。追踪数据流，测试假设，3 次修复失败后停止。 |\n| `\u002Fdesign-review` | **会代码的设计师** | 与 \u002Fplan-design-review 相同的审计，然后修复发现的问题。原子提交，前后截图对比。 |\n| `\u002Fqa` | **质量保证主管** | 测试你的应用，发现 bug，通过原子提交修复它们，重新验证。为每个修复自动生成回归测试。 |\n| `\u002Fqa-only` | **质量保证报告员** | 与 \u002Fqa 方法相同，但仅报告。纯 bug 报告，无代码更改。 |\n| `\u002Fcso` | **首席安全官** | OWASP Top 10 (十大安全风险) + STRIDE (威胁建模) 威胁模型。零噪音：17 个误报排除，8\u002F10+ 置信度门槛，独立发现验证。每个发现都包含具体的利用场景。 |\n| `\u002Fship` | **发布工程师** | 同步主分支，运行测试，审计覆盖率，推送，打开拉取请求 (PR)。如果你没有测试框架，它会引导初始化。 |\n| `\u002Fland-and-deploy` | **发布工程师** | 合并 PR，等待 CI 并部署，验证生产环境健康。一个命令从“已批准”到“生产环境验证”。 |\n| `\u002Fcanary` | **站点可靠性工程师 (SRE)** | 部署后监控循环。监视控制台错误、性能回归和页面故障。 |\n| `\u002Fbenchmark` | **性能工程师** | 基线页面加载时间、Core Web Vitals (核心网页指标) 和资源大小。在每个 PR 上比较前后变化。 |\n| `\u002Fdocument-release` | **技术文档工程师** | 更新所有项目文档以匹配你刚刚发布的内容。自动捕获过时的 README。 |\n| `\u002Fretro` | **工程经理** | 团队感知的每周复盘。每人细分、发布条纹、测试健康趋势、增长机会。`\u002Fretro global` 跨所有项目和 AI 工具（Claude Code, Codex, Gemini）运行。 |\n| `\u002Fbrowse` | **质量保证工程师** | 给智能体装上眼睛。真实的 Chromium (浏览器内核) 浏览器，真实的点击，真实的截图。每个命令约 100ms。`$B connect` 将你真实的 Chrome 作为有头窗口 (可见窗口) 启动——实时观看每个动作。 |\n| `\u002Fsetup-browser-cookies` | **会话管理器** | 将你真实浏览器（Chrome, Arc, Brave, Edge）的 Cookie 导入无头 (Headless) 会话。测试认证页面。 |\n| `\u002Fautoplan` | **审查流水线** | 一个命令，完全审查的计划。自动运行 CEO → 设计 → 工程审查，编码决策原则。仅表面化品味决策供你批准。 |\n\n### 强力工具\n\n| 技能 | 职责 |\n|-------|-------------|\n| `\u002Fcodex` | **第二意见** — 来自 OpenAI Codex 命令行界面 (CLI) 的独立代码审查。三种模式：审查（通过\u002F失败门槛）、对抗性挑战和开放咨询。当 `\u002Freview` 和 `\u002Fcodex` 都运行后进行跨模型分析。 |\n| `\u002Fcareful` | **安全护栏** — 在执行破坏性命令（rm -rf, DROP TABLE, 强制推送）前警告。说\"be careful\"激活。可覆盖任何警告。 |\n| `\u002Ffreeze` | **编辑锁定** — 将文件编辑限制在一个目录内。防止调试时意外更改范围外的内容。 |\n| `\u002Fguard` | **全面安全** — `\u002Fcareful` + `\u002Ffreeze` 合二为一。生产环境 (Prod) 工作的最大安全性。 |\n| `\u002Funfreeze` | **解锁** — 移除 `\u002Ffreeze` 边界。 |\n| `\u002Fsetup-deploy` | **部署配置器** — `\u002Fland-and-deploy` 的一次性设置。检测你的平台、生产环境 URL 和部署命令。 |\n| `\u002Fgstack-upgrade` | **自我更新器** — 升级 gstack 到最新版本。检测全局与本地依赖安装，同步两者，显示变更内容。 |\n\n**[深入了解每个技能的示例和理念 →](docs\u002Fskills.md)**\n\n## 并行 sprint (冲刺)\n\ngstack 在一个 sprint (冲刺) 中表现良好。当十个同时运行时才真正有趣。\n\n**设计是核心。** `\u002Fdesign-consultation` 不仅仅是挑选字体。它研究你所在领域的现状，提出安全的选择和创造性的风险方案，生成你实际产品的逼真 mockup (原型)，并编写 `DESIGN.md` — 然后 `\u002Fdesign-review` 和 `\u002Fplan-eng-review` 会阅读你所选择的内容。设计决策流经整个系统。\n\n**`\u002Fqa` 是一个巨大的 unlock (解锁)。** 它让我从 6 个并行 workers (工作器) 增加到 12 个。Claude Code 说 *\"I SEE THE ISSUE\"* 然后实际上修复它，生成 regression test (回归测试)，并验证修复 — 这改变了我的工作方式。agent (智能体) 现在有了眼睛。\n\n**智能审查路由。** 就像在一家运营良好的初创公司：CEO 不必查看 infra (基础设施) 的 bug 修复，后端更改不需要设计审查。gstack 跟踪运行了哪些审查，弄清楚什么是合适的，然后只做聪明的事。Review Readiness Dashboard (审查就绪仪表板) 告诉你在发布 (ship) 前的状态。\n\n**测试一切。** 如果你的项目没有测试框架，`\u002Fship` 会从头 bootstraps (引导启动) 测试框架。每次 `\u002Fship` 运行都会产生 coverage audit (覆盖率审计)。每次 `\u002Fqa` bug 修复都会生成回归测试。100% 测试覆盖率是目标 — 测试让 vibe coding (凭感觉编码) 变得安全，而不是 yolo coding (孤注一掷编码)。\n\n**`\u002Fdocument-release` 是你从未拥有过的工程师。** 它读取项目中的每个文档文件，交叉引用 diff (差异)，并更新所有漂移的内容。README、ARCHITECTURE、CONTRIBUTING、CLAUDE.md、TODOS — 全部自动保持最新。现在 `\u002Fship` 会自动调用它 — 文档保持最新而无需额外命令。\n\n**真实浏览器模式。** `$B connect` 启动你实际的 Chrome 作为由 Playwright 控制的 headed window (有头窗口)。你实时观看 Claude 点击、填写和导航 — 同一窗口，同一屏幕。顶部边缘微弱的绿色闪烁告诉你哪个 Chrome 窗口由 gstack 控制。所有现有的浏览命令保持不变。`$B disconnect` 返回 headless (无头模式)。Chrome 扩展侧边栏 (Side Panel) 显示每个命令的实时活动反馈，以及一个聊天侧边栏，你可以在其中指导 Claude。这是 co-presence (共在) — Claude 不是远程控制隐藏的浏览器，它就在同一个驾驶舱里坐在你旁边。\n\n**侧边栏 agent (智能体) — 你的 AI 浏览器助手。** 在 Chrome 侧边栏中输入自然语言指令，子 Claude 实例执行它们。\"Navigate to the settings page and screenshot it.\" \"Fill out this form with test data.\" \"Go through every item in this list and extract the prices.\" 每个任务最多 5 分钟。侧边栏 agent 在隔离 session (会话) 中运行，所以它不会干扰你的主 Claude Code 窗口。就像在浏览器中多了一双手。\n\n**个人自动化。** 侧边栏 agent 不仅用于开发工作流。例如：\"Browse my kid's school parent portal and add all the other parents' names, phone numbers, and photos to my Google Contacts.\" 两种认证方式：(1) 在有头浏览器中登录一次 — 你的 session (会话) 持久化，或 (2) 运行 `\u002Fsetup-browser-cookies` 从你真实的 Chrome 导入 cookies (Cookie)。一旦认证，Claude 导航目录，提取数据，并创建联系人。\n\n**当 AI 卡住时的浏览器 handoff (交接)。** 遇到 CAPTCHA (验证码)、auth wall (认证墙) 或 MFA (多因素认证) 提示？`$B handoff` 在完全相同的页面打开一个可见的 Chrome，所有你的 cookies 和标签页完好无损。解决问题，告诉 Claude 你完成了，`$B resume` 接着断点继续。agent 甚至在连续 3 次失败后自动建议。\n\n**多 AI 第二意见。** `\u002Fcodex` 从 OpenAI 的 Codex CLI 获得独立审查 — 一个完全不同的 AI 查看相同的 diff。三种模式：带有 pass\u002Ffail gate (通过\u002F失败网关) 的代码审查，主动尝试破坏你代码的对抗性挑战，以及具有会话连续性的开放咨询。当 `\u002Freview` (Claude) 和 `\u002Fcodex` (OpenAI) 都审查了同一分支时，你会获得跨模型分析，显示哪些发现重叠，哪些是每个模型独有的。\n\n**按需安全护栏。** 说 \"be careful\"，`\u002Fcareful` 会在任何破坏性命令前警告 — rm -rf、DROP TABLE、force-push、git reset --hard。`\u002Ffreeze` 在调试时将编辑锁定到一个目录，所以 Claude 不能意外\"修复\"无关代码。`\u002Fguard` 激活两者。`\u002Finvestigate` 自动冻结到正在调查的模块。\n\n**主动技能建议。** gstack 注意到你处于什么阶段 — 头脑风暴、审查、调试、测试 — 并建议合适的技能。不喜欢？说 \"stop suggesting\"，它会跨会话记住。\n\n## 10-15 个并行 sprint (冲刺)\n\ngstack 在一个 sprint 中很强大。当十个同时运行时它具有变革性。\n\n[Conductor](https:\u002F\u002Fconductor.build) 并行运行多个 Claude Code 会话 — 每个都在其自己的隔离工作区中。一个会话在新想法上运行 `\u002Foffice-hours`，另一个在 PR (拉取请求) 上做 `\u002Freview`，第三个实现功能，第四个在 staging (预发布环境) 上运行 `\u002Fqa`，还有六个在其他分支上。全部同时。我经常运行 10-15 个并行 sprint — 这是目前的实际最大值。\n\nsprint 结构使并行工作成为可能。没有流程，十个 agent 就是十个混乱源。有了流程 — 思考、计划、构建、审查、测试、发布 — 每个 agent 都知道确切该做什么以及何时停止。你像 CEO 管理团队一样管理它们：检查重要的决策，让其余的运行。\n\n---\n\n免费，MIT 许可，开源。无高级 tier (层级)，无等待名单。\n\n我开源了我构建软件的方式。你可以 fork (复刻) 它并使其成为你自己的。\n\n> **我们正在招聘。** 想要每天发布 10K+ LOC (代码行) 并帮助强化 gstack？\n> 来 YC 工作 — [ycombinator.com\u002Fsoftware](https:\u002F\u002Fycombinator.com\u002Fsoftware)\n> 极具竞争力的薪资和股权。旧金山，Dogpatch 区。\n\n## 文档\n\n| 文档 | 涵盖内容 |\n|-----|---------------|\n| [技能深入探讨](docs\u002Fskills.md) | 每个技能的哲学、示例和工作流（包括 Greptile 集成） |\n| [构建者理念](ETHOS.md) | 构建者哲学：Boil the Lake (煮湖\u002F宏大目标), Search Before Building (构建前搜索), three layers of knowledge (三层知识) |\n| [架构](ARCHITECTURE.md) | 设计决策和系统内部 |\n| [浏览器参考](BROWSER.md) | `\u002Fbrowse` 的完整命令参考 |\n| [贡献指南](CONTRIBUTING.md) | 开发设置、测试、贡献者模式和开发模式 |\n| [变更日志](CHANGELOG.md) | 每个版本的新内容 |\n\n## 隐私与遥测 (Telemetry)\n\ngstack 包含**选择性加入 (opt-in)** 的使用遥测数据，以帮助改进项目。具体情况如下：\n\n- **默认为关闭。** 除非你明确同意，否则不会发送任何数据。\n- **首次运行时，** gstack 会询问你是否愿意分享匿名使用数据。你可以拒绝。\n- **发送的内容（如果你选择加入）：** 技能名称、持续时间、成功\u002F失败、gstack 版本、操作系统 (OS)。仅此而已。\n- **绝不发送的内容：** 代码、文件路径、仓库名称、分支名称、提示词 (prompts) 或任何用户生成的内容。\n- **随时更改：** `gstack-config set telemetry off` 可立即禁用所有功能。\n\n数据存储在 [Supabase](https:\u002F\u002Fsupabase.com)（开源 Firebase 替代品）中。模式 (schema) 位于 [`supabase\u002Fmigrations\u002F`](supabase\u002Fmigrations\u002F) — 你可以确切地验证收集了什么内容。仓库中的 Supabase 可发布密钥是一个公钥（类似于 Firebase API 密钥）— 行级安全策略 (Row-level security policies) 拒绝所有直接访问。遥测数据流经经过验证的边缘函数 (edge functions)，这些函数强制执行模式检查、事件类型允许列表和字段长度限制。\n\n**本地分析始终可用。** 运行 `gstack-analytics` 即可查看本地 JSONL 文件中的个人使用仪表板 — 无需远程数据。\n\n## 故障排除\n\n**技能未显示？** `cd ~\u002F.claude\u002Fskills\u002Fgstack && .\u002Fsetup`\n\n**`\u002Fbrowse` 失败？** `cd ~\u002F.claude\u002Fskills\u002Fgstack && bun install && bun run build`\n\n**安装过时？** 运行 `\u002Fgstack-upgrade` — 或在 `~\u002F.gstack\u002Fconfig.yaml` 中设置 `auto_upgrade: true`\n\n**Codex 显示 \"Skipped loading skill(s) due to invalid SKILL.md\"？** 你的 Codex 技能描述已过时。解决方法：`cd ~\u002F.codex\u002Fskills\u002Fgstack && git pull && .\u002Fsetup --host codex` — 或者对于仓库本地安装：`cd \"$(readlink -f .agents\u002Fskills\u002Fgstack)\" && git pull && .\u002Fsetup --host codex`\n\n**Windows 用户：** gstack 可通过 Git Bash 或 WSL 在 Windows 11 上运行。除了 Bun 外还需要 Node.js — Bun 在 Windows 上与 Playwright 的管道传输 (pipe transport) 存在已知问题 ([bun#4253](https:\u002F\u002Fgithub.com\u002Foven-sh\u002Fbun\u002Fissues\u002F4253))。浏览服务器会自动回退到 Node.js。确保 `bun` 和 `node` 都在你的 PATH 环境变量中。\n\n**Claude 说它看不到技能？** 确保你项目的 `CLAUDE.md` 包含 gstack 部分。添加以下内容：\n\n```\n## gstack\nUse \u002Fbrowse from gstack for all web browsing. Never use mcp__claude-in-chrome__* tools.\nAvailable skills: \u002Foffice-hours, \u002Fplan-ceo-review, \u002Fplan-eng-review, \u002Fplan-design-review,\n\u002Fdesign-consultation, \u002Freview, \u002Fship, \u002Fland-and-deploy, \u002Fcanary, \u002Fbenchmark, \u002Fbrowse,\n\u002Fqa, \u002Fqa-only, \u002Fdesign-review, \u002Fsetup-browser-cookies, \u002Fsetup-deploy, \u002Fretro,\n\u002Finvestigate, \u002Fdocument-release, \u002Fcodex, \u002Fcso, \u002Fautoplan, \u002Fcareful, \u002Ffreeze, \u002Fguard,\n\u002Funfreeze, \u002Fgstack-upgrade.\n```\n\n## 许可证\n\nMIT。永久免费。去构建点什么吧。","# gstack 快速上手指南\n\ngstack 是由 Y Combinator CEO Garry Tan 开源的 AI 工程化工作流工具。它将 Claude Code 转化为一个虚拟工程团队（包含 CEO、工程经理、设计师、QA 等角色），通过标准化的 Slash 命令实现从构思到部署的全流程自动化。\n\n## 环境准备\n\n在使用 gstack 之前，请确保您的开发环境满足以下要求：\n\n- **AI 代理**：[Claude Code](https:\u002F\u002Fdocs.anthropic.com\u002Fen\u002Fdocs\u002Fclaude-code)（核心依赖）\n- **版本控制**：[Git](https:\u002F\u002Fgit-scm.com\u002F)\n- **运行时环境**：[Bun](https:\u002F\u002Fbun.sh\u002F) v1.0+\n- **其他**：[Node.js](https:\u002F\u002Fnodejs.org\u002F)（仅 Windows 系统需要）\n- **网络**：需能够正常访问 GitHub 及 Claude 服务\n\n## 安装步骤\n\ngstack 支持个人全局安装和项目局部安装。建议先进行个人安装，再根据需要添加到具体项目。\n\n### 1. 个人全局安装（30 秒）\n\n在终端中执行以下命令，将 gstack 安装到 Claude 技能目录并完成初始化：\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack.git ~\u002F.claude\u002Fskills\u002Fgstack && cd ~\u002F.claude\u002Fskills\u002Fgstack && .\u002Fsetup\n```\n\n**配置 CLAUDE.md：**\n安装完成后，需确保您的 `CLAUDE.md` 文件中包含 gstack 配置段。如果使用 Claude Code，可粘贴以下指令让 AI 自动配置：\n\n> Add a \"gstack\" section to CLAUDE.md that says to use the \u002Fbrowse skill from gstack for all web browsing, never use mcp__claude-in-chrome__* tools, and lists the available skills: \u002Foffice-hours, \u002Fplan-ceo-review, \u002Fplan-eng-review, \u002Fplan-design-review, \u002Fdesign-consultation, \u002Freview, \u002Fship, \u002Fland-and-deploy, \u002Fcanary, \u002Fbenchmark, \u002Fbrowse, \u002Fqa, \u002Fqa-only, \u002Fdesign-review, \u002Fsetup-browser-cookies, \u002Fsetup-deploy, \u002Fretro, \u002Finvestigate, \u002Fdocument-release, \u002Fcodex, \u002Fcso, \u002Fautoplan, \u002Fcareful, \u002Ffreeze, \u002Fguard, \u002Funfreeze, \u002Fgstack-upgrade.\n\n### 2. 添加到项目仓库（可选）\n\n若希望团队成员共享该配置，可将 gstack 复制到当前项目目录：\n\n```bash\ncp -Rf ~\u002F.claude\u002Fskills\u002Fgstack .claude\u002Fskills\u002Fgstack && rm -rf .claude\u002Fskills\u002Fgstack\u002F.git && cd .claude\u002Fskills\u002Fgstack && .\u002Fsetup\n```\n\n随后在项目根目录的 `CLAUDE.md` 中添加相同的 gstack 配置段。\n\n### 3. 其他 AI 代理支持\n\ngstack 同样支持 Codex、Gemini CLI 或 Cursor 等支持 [SKILL.md 标准](https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fclaude-code) 的代理。\n\n```bash\n# 安装到当前仓库\ngit clone https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack.git .agents\u002Fskills\u002Fgstack\ncd .agents\u002Fskills\u002Fgstack && .\u002Fsetup --host codex\n\n# 或自动检测已安装的代理\ngit clone https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack.git ~\u002Fgstack\ncd ~\u002Fgstack && .\u002Fsetup --host auto\n```\n\n## 基本使用\n\ngstack 的核心是一个标准化的研发冲刺流程：**思考 → 计划 → 构建 → 审查 → 测试 → 部署 → 复盘**。\n\n### 1. 启动流程\n在 Claude Code 中输入 `\u002Foffice-hours` 开始描述你的构建想法。AI 会像 YC 导师一样反问痛点，挑战前提，并生成设计文档。\n\n```\nYou:    I want to build a daily briefing app for my calendar.\nYou:    \u002Foffice-hours\nClaude: [asks about the pain — specific examples, not hypotheticals]\n```\n\n### 2. 核心命令工作流\n\n按照以下顺序使用 Slash 命令，即可驱动虚拟团队完成开发：\n\n| 步骤 | 命令 | 角色 | 作用 |\n| :--- | :--- | :--- | :--- |\n| **计划** | `\u002Fplan-ceo-review` | CEO | 重新审视问题，确定产品范围与优先级 |\n| **架构** | `\u002Fplan-eng-review` | 工程经理 | 锁定架构、数据流、边界情况与测试计划 |\n| **构建** | (自动) | 工程师 | 退出计划模式后，AI 自动编写代码 |\n| **审查** | `\u002Freview` | 资深工程师 | 查找潜在 Bug，自动修复明显问题 |\n| **测试** | `\u002Fqa \u003CURL>` | QA 主管 | 打开真实浏览器测试流程，发现并修复 Bug |\n| **部署** | `\u002Fship` | 发布工程师 | 同步主干、运行测试、提交 PR |\n| **上线** | `\u002Fland-and-deploy` | 发布工程师 | 合并 PR 并验证生产环境健康度 |\n\n### 3. 使用示例\n\n以下是一个完整的功能开发对话示例：\n\n```\nYou:    \u002Fplan-ceo-review\n        [reads the design doc, challenges scope, runs 10-section review]\n\nYou:    \u002Fplan-eng-review\n        [ASCII diagrams for data flow, state machines, error paths]\n\nYou:    Approve plan. Exit plan mode.\n        [writes 2,400 lines across 11 files. ~8 minutes.]\n\nYou:    \u002Freview\n        [AUTO-FIXED] 2 issues. [ASK] Race condition → you approve fix.\n\nYou:    \u002Fqa https:\u002F\u002Fstaging.myapp.com\n        [opens real browser, clicks through flows, finds and fixes a bug]\n\nYou:    \u002Fship\n        Tests: 42 → 51 (+9 new). PR: github.com\u002Fyou\u002Fapp\u002Fpull\u002F42\n```\n\n### 4. 其他实用命令\n\n- `\u002Fretro`：生成团队感知的每周复盘报告。\n- `\u002Fcso`：首席安全官，运行 OWASP Top 10 + STRIDE 威胁模型审计。\n- `\u002Fbenchmark`：性能工程师，对比 PR 前后的页面加载时间与 Core Web Vitals。\n- `\u002Fdocument-release`：自动更新项目文档以匹配最新发布的代码。\n\n> **提示**：所有技能均基于 Markdown 和 Slash 命令，无需额外背景进程。如遇技能无法使用，请运行 `cd .claude\u002Fskills\u002Fgstack && .\u002Fsetup` 重新构建二进制文件并注册技能。","一位技术创始人正在独自开发 SaaS 平台的核心计费模块，同时需要处理公司日常运营事务，时间极其碎片化。\n\n### 没有 gstack 时\n- 需要在产品决策、代码编写和测试之间频繁切换，精力分散导致效率低下。\n- 缺乏专人代码审查，逻辑漏洞难以自查，上线后偶发严重 Bug，修复成本极高。\n- 手动测试流程繁琐，容易遗漏边缘场景，每次发布都提心吊胆。\n- 文档更新滞后于代码迭代，团队成员难以跟进最新业务逻辑。\n- 独自承担所有角色，长期高压下难以保证交付节奏。\n\n### 使用 gstack 后\n- gstack 的 `\u002Fplan-ceo-review` 辅助产品决策，`\u002Freview` 自动完成严格代码审查，弥补单人视角盲区。\n- 通过 `\u002Fqa` 命令在真实浏览器中自动化测试，提前拦截生产环境隐患，发布信心大幅提升。\n- 内置的 `\u002Fdocument-release` 同步生成发布文档，保持文档与代码实时一致，降低沟通成本。\n- gstack 扮演了工程师、QA 和发布经理的角色，创始人只需关注核心架构与关键决策。\n- 依托预设的 15 个专业工具链，单人也能维持高标准工程规范，交付速度提升数倍。\n\ngstack 让单人开发者拥有完整工程团队的执行力，显著提升交付质量与速度。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgarrytan_gstack_166fb3b0.png","garrytan","Garry Tan","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fgarrytan_b8e4e455.jpg","Writes software, dreams. ","Y Combinator","San Francisco, CA","garrytan@gmail.com","https:\u002F\u002Fblog.garrytan.com","https:\u002F\u002Fgithub.com\u002Fgarrytan",[85,89,93,97,101,105],{"name":86,"color":87,"percentage":88},"TypeScript","#3178c6",72.2,{"name":90,"color":91,"percentage":92},"Go Template","#00ADD8",21,{"name":94,"color":95,"percentage":96},"Shell","#89e051",3.8,{"name":98,"color":99,"percentage":100},"JavaScript","#f1e05a",1.9,{"name":102,"color":103,"percentage":104},"CSS","#663399",0.9,{"name":106,"color":107,"percentage":108},"HTML","#e34c26",0.3,64261,8734,"2026-04-05T11:08:43","MIT","Linux, macOS, Windows","不需要 (基于云端 AI 服务)","未说明",{"notes":117,"python":118,"dependencies":119},"该工具是用于 AI 代理（如 Claude Code）的技能集，非本地机器学习模型。需预先安装并配置 Claude Code。安装通过 git clone 及.\u002Fsetup 脚本完成，技能文件存储于.claude\u002Fskills 或.agents\u002Fskills 目录。支持多种 AI 代理平台（Claude Code, Codex, Gemini, Cursor）。","未说明 (基于 Bun\u002FNode.js)",[120,121,122,123],"Claude Code","Git","Bun v1.0+","Node.js (Windows only)",[15,45],null,298,"2026-03-27T02:34:52.378252","2026-04-06T02:31:30.272389",[130,135,139,143,147,152],{"id":131,"question_zh":132,"answer_zh":133,"source_url":134},1,"使用 Codex CLI 时提示 SKILL.md 描述超过 1024 字符限制错误怎么办？","该问题已在 #346 (Wave 2 Codex Compatibility) 中修复。请确保更新 gstack 到最新版本。如果问题仍然存在，可能是安装路径或兼容性问题，建议重新运行 setup 脚本。维护者确认修复后，重启 Codex 即可恢复正常。","https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack\u002Fissues\u002F230",{"id":23,"question_zh":136,"answer_zh":137,"source_url":138},"如何手动解决 Codex 描述超过 1024 字符的错误（临时方案）？","可以手动缩短 `SKILL.md.tmpl` 中的 `description` 字段至 1024 字符以内。社区用户提供的压缩后描述示例如下（约 559 字符）：\n\n```\ndescription: |\n  Fast headless browser for QA and dogfooding: navigate URLs, interact with elements, diff\u002Fverify page state, take annotated screenshots, test responsive layouts\u002Fforms\u002Fuploads\u002Fdialogs, and assert UI state (~100ms\u002Fcommand). Use for feature testing, deployment checks, end-to-end dogfooding, and evidence-backed bug reports.\n  \n  gstack also includes workflow skills. Suggest when relevant: \u002Foffice-hours, \u002Fplan-ceo-review, \u002Fplan-eng-review, \u002Fplan-design-review, \u002Fdesign-consultation, \u002Finvestigate, \u002Fqa, \u002Freview, \u002Fdesign-review, \u002Fship, \u002Fdocument-release, \u002Fretro, \u002Fcodex, \u002Fcareful, \u002Ffreeze, \u002Fguard, \u002Funfreeze, \u002Fgstack-upgrade.\n```\n\n修改后重新生成 `SKILL.md` 并重启 Codex。","https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack\u002Fissues\u002F281",{"id":10,"question_zh":140,"answer_zh":141,"source_url":142},"gstack 支持 Windows 系统吗？","支持。Windows 支持已在 v0.9.3.0 版本中发布。Windows 11 用户可通过 Git Bash 或 WSL 使用。设置过程不再挂起，browse 服务器会自动回退到 Node.js。注意：在 Windows 上需要同时安装 Node.js 和 Bun。","https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack\u002Fissues\u002F69",{"id":144,"question_zh":145,"answer_zh":146,"source_url":142},4,"在 Windows 上 browse 功能无法正常工作如何解决？","建议在 WSL 中运行 browse 服务器。具体步骤：\n1. 在 WSL 终端启动服务：\n```bash\nwsl\ncd ~\u002F.claude\u002Fskills\u002Fgstack\u002Fbrowse\nbun run src\u002Fserver.ts\n```\n2. 服务器会输出端口和 Token，例如 `http:\u002F\u002F0.0.0.0:27703`。\n3. 从 Windows 可通过 `\\\\wsl$\\Ubuntu\\home\\...\\.gstack\\browse.json` 访问状态文件，或使用 curl 发送命令到 WSL 暴露的端口进行测试。",{"id":148,"question_zh":149,"answer_zh":150,"source_url":151},5,"提示 `\u002Fgstack-upgrade` 不是 recognized skill 怎么办？","当前版本可能缺少该技能的文档。建议使用以下命令重新添加技能：\n```bash\nnpx skills add garrytan\u002Fgstack\n```\n或者指向仓库获取最新版本。如果仍然无效，可以尝试手动更新方法。","https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgstack\u002Fissues\u002F68",{"id":153,"question_zh":154,"answer_zh":155,"source_url":151},6,"如何手动更新 gstack 到最新版本？","可以通过 Git 命令手动拉取最新代码并重置。运行以下命令：\n```bash\ncd ~\u002F.claude\u002Fskills\u002Fgstack && git fetch origin && git reset --hard origin\u002Fmain && .\u002Fsetup\n```\n如果项目也在 `.claude\u002Fskills\u002Fgstack`，需同步更新并清理旧技能文件：\n```bash\nfor s in browse plan-ceo-review plan-eng-review review ship retro qa setup-browser-cookies; do rm -f .claude\u002Fskills\u002F$s; done && rm -rf .claude\u002Fskills\u002Fgstack && cp -Rf ~\u002F.claude\u002Fskills\u002Fgstack .claude\u002Fskills\u002Fgstack && rm -rf .claude\u002Fskills\u002Fgstack\u002F.git && cd .claude\u002Fskills\u002Fgstack && .\u002Fsetup\n```",[]]