[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-sdi2200262--agentic-project-management":3,"tool-sdi2200262--agentic-project-management":62},[4,18,28,37,45,53],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":17},4358,"openclaw","openclaw\u002Fopenclaw","OpenClaw 是一款专为个人打造的本地化 AI 助手，旨在让你在自己的设备上拥有完全可控的智能伙伴。它打破了传统 AI 助手局限于特定网页或应用的束缚，能够直接接入你日常使用的各类通讯渠道，包括微信、WhatsApp、Telegram、Discord、iMessage 等数十种平台。无论你在哪个聊天软件中发送消息，OpenClaw 都能即时响应，甚至支持在 macOS、iOS 和 Android 设备上进行语音交互，并提供实时的画布渲染功能供你操控。\n\n这款工具主要解决了用户对数据隐私、响应速度以及“始终在线”体验的需求。通过将 AI 部署在本地，用户无需依赖云端服务即可享受快速、私密的智能辅助，真正实现了“你的数据，你做主”。其独特的技术亮点在于强大的网关架构，将控制平面与核心助手分离，确保跨平台通信的流畅性与扩展性。\n\nOpenClaw 非常适合希望构建个性化工作流的技术爱好者、开发者，以及注重隐私保护且不愿被单一生态绑定的普通用户。只要具备基础的终端操作能力（支持 macOS、Linux 及 Windows WSL2），即可通过简单的命令行引导完成部署。如果你渴望拥有一个懂你",349277,3,"2026-04-06T06:32:30",[13,14,15,16],"Agent","开发框架","图像","数据工具","ready",{"id":19,"name":20,"github_repo":21,"description_zh":22,"stars":23,"difficulty_score":24,"last_commit_at":25,"category_tags":26,"status":17},9989,"n8n","n8n-io\u002Fn8n","n8n 是一款面向技术团队的公平代码（fair-code）工作流自动化平台，旨在让用户在享受低代码快速构建便利的同时，保留编写自定义代码的灵活性。它主要解决了传统自动化工具要么过于封闭难以扩展、要么完全依赖手写代码效率低下的痛点，帮助用户轻松连接 400 多种应用与服务，实现复杂业务流程的自动化。\n\nn8n 特别适合开发者、工程师以及具备一定技术背景的业务人员使用。其核心亮点在于“按需编码”：既可以通过直观的可视化界面拖拽节点搭建流程，也能随时插入 JavaScript 或 Python 代码、调用 npm 包来处理复杂逻辑。此外，n8n 原生集成了基于 LangChain 的 AI 能力，支持用户利用自有数据和模型构建智能体工作流。在部署方面，n8n 提供极高的自由度，支持完全自托管以保障数据隐私和控制权，也提供云端服务选项。凭借活跃的社区生态和数百个现成模板，n8n 让构建强大且可控的自动化系统变得简单高效。",184740,2,"2026-04-19T23:22:26",[16,14,13,15,27],"插件",{"id":29,"name":30,"github_repo":31,"description_zh":32,"stars":33,"difficulty_score":10,"last_commit_at":34,"category_tags":35,"status":17},10095,"AutoGPT","Significant-Gravitas\u002FAutoGPT","AutoGPT 是一个旨在让每个人都能轻松使用和构建 AI 的强大平台，核心功能是帮助用户创建、部署和管理能够自动执行复杂任务的连续型 AI 智能体。它解决了传统 AI 应用中需要频繁人工干预、难以自动化长流程工作的痛点，让用户只需设定目标，AI 即可自主规划步骤、调用工具并持续运行直至完成任务。\n\n无论是开发者、研究人员，还是希望提升工作效率的普通用户，都能从 AutoGPT 中受益。开发者可利用其低代码界面快速定制专属智能体；研究人员能基于开源架构探索多智能体协作机制；而非技术背景用户也可直接选用预置的智能体模板，立即投入实际工作场景。\n\nAutoGPT 的技术亮点在于其模块化“积木式”工作流设计——用户通过连接功能块即可构建复杂逻辑，每个块负责单一动作，灵活且易于调试。同时，平台支持本地自托管与云端部署两种模式，兼顾数据隐私与使用便捷性。配合完善的文档和一键安装脚本，即使是初次接触的用户也能在几分钟内启动自己的第一个 AI 智能体。AutoGPT 正致力于降低 AI 应用门槛，让人人都能成为 AI 的创造者与受益者。",183572,"2026-04-20T04:47:55",[13,36,27,14,15],"语言模型",{"id":38,"name":39,"github_repo":40,"description_zh":41,"stars":42,"difficulty_score":10,"last_commit_at":43,"category_tags":44,"status":17},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,"2026-04-05T11:01:52",[14,15,13],{"id":46,"name":47,"github_repo":48,"description_zh":49,"stars":50,"difficulty_score":24,"last_commit_at":51,"category_tags":52,"status":17},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",161147,"2026-04-19T23:31:47",[14,13,36],{"id":54,"name":55,"github_repo":56,"description_zh":57,"stars":58,"difficulty_score":59,"last_commit_at":60,"category_tags":61,"status":17},8272,"opencode","anomalyco\u002Fopencode","OpenCode 是一款开源的 AI 编程助手（Coding Agent），旨在像一位智能搭档一样融入您的开发流程。它不仅仅是一个代码补全插件，而是一个能够理解项目上下文、自主规划任务并执行复杂编码操作的智能体。无论是生成全新功能、重构现有代码，还是排查难以定位的 Bug，OpenCode 都能通过自然语言交互高效完成，显著减少开发者在重复性劳动和上下文切换上的时间消耗。\n\n这款工具专为软件开发者、工程师及技术研究人员设计，特别适合希望利用大模型能力来提升编码效率、加速原型开发或处理遗留代码维护的专业人群。其核心亮点在于完全开源的架构，这意味着用户可以审查代码逻辑、自定义行为策略，甚至私有化部署以保障数据安全，彻底打破了传统闭源 AI 助手的“黑盒”限制。\n\n在技术体验上，OpenCode 提供了灵活的终端界面（Terminal UI）和正在测试中的桌面应用程序，支持 macOS、Windows 及 Linux 全平台。它兼容多种包管理工具，安装便捷，并能无缝集成到现有的开发环境中。无论您是追求极致控制权的资深极客，还是渴望提升产出的独立开发者，OpenCode 都提供了一个透明、可信",144296,1,"2026-04-16T14:50:03",[13,27],{"id":63,"github_repo":64,"name":65,"description_en":66,"description_zh":67,"ai_summary_zh":67,"readme_en":68,"readme_zh":69,"quickstart_zh":70,"use_case_zh":71,"hero_image_url":72,"owner_login":73,"owner_name":74,"owner_avatar_url":75,"owner_bio":76,"owner_company":77,"owner_location":78,"owner_email":77,"owner_twitter":77,"owner_website":77,"owner_url":79,"languages":80,"stars":85,"forks":86,"last_commit_at":87,"license":88,"difficulty_score":24,"env_os":89,"env_gpu":89,"env_ram":89,"env_deps":90,"category_tags":95,"github_topics":77,"view_count":24,"oss_zip_url":77,"oss_zip_packed_at":77,"status":17,"created_at":96,"updated_at":97,"faqs":98,"releases":99},9012,"sdi2200262\u002Fagentic-project-management","agentic-project-management","A framework for managing complex projects with spec-driven multi-agent workflows.","Agentic Project Management（APM）是一款专为管理复杂软件项目而设计的开源框架，它通过“规范驱动”的多智能体协作模式，让 AI 团队像人类一样分工合作。\n\n在传统的单一大模型对话中，随着项目规模扩大，AI 往往因上下文过长而迷失方向，导致需求遗漏、代码质量下降甚至产生幻觉。APM 巧妙地将工作流拆解为三个专职角色：**规划师**负责拆解需求并制定规格与规则；**管理员**统筹任务分配与进度审查；**执行者**则专注于在具体上下文中完成编码任务。项目状态被存储在结构化文件中而非依赖对话记忆，确保即使会话中断或切换，核心信息也能无缝传承。\n\n这一架构特别适合需要开发大型应用、维护复杂代码库的开发者及技术团队。其独特亮点在于“上下文隔离”与“人工介入机制”：每个智能体仅获取必要信息以保持专注，而用户作为协调者全程可见并掌控每一步流转，既避免了黑盒操作，又兼容 Claude Code、Cursor、GitHub Copilot 等主流 AI 编程助手。通过 APM，你可以将混乱的长对话转化为井然有序的工程流水线，让 AI 真正胜任长期且复杂的开发任务。","# Agentic Project Management (APM)\n\n[![License: MPL-2.0](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MPL_2.0-brightgreen.svg)](https:\u002F\u002Fopensource.org\u002Flicenses\u002FMPL-2.0) [![npm](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fv\u002Fagentic-pm)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002Fagentic-pm) [![GitHub Release](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fv\u002Frelease\u002Fsdi2200262\u002Fagentic-project-management)](https:\u002F\u002Fgithub.com\u002Fsdi2200262\u002Fagentic-project-management\u002Freleases)\n\n*Manage complex projects with a team of AI agents, smoothly and efficiently.*\n\n## What is APM?\n\nAPM is an open-source framework for managing ambitious software projects with AI assistants. Instead of working in a single, increasingly chaotic chat, APM structures your work into a coordinated system where different AI agents handle planning, coordination, and execution as a team.\n\nAs conversations grow, AI context degrades. The assistant loses track of requirements, produces bad code, and hallucinates details. For substantial projects, this makes sustained progress nearly impossible.\n\nTo address this, APM coordinates three specialized agent types, each operating in its own context with only the information it needs:\n\n- **Planner** - Conducts structured project discovery and decomposes requirements into three planning documents: a Spec (what to build), a Plan (how work is organized), and Rules (how work is performed).\n- **Manager** - Coordinates execution by assigning Tasks to Workers, reviewing completed work, and maintaining project state. Operates on execution summaries rather than raw code.\n- **Workers** - Execute specific Tasks with tightly scoped context. Each Worker receives self-contained Task Prompts with everything it needs, executes, validates, and reports back.\n\nProject state lives in structured files outside any agent's context. When a conversation ends or an agent reaches its limits, a Handoff transfers working knowledge to a fresh instance as needed. This also allows completed APM sessions to be archived and their context carried forward to new ones.\n\nYou mediate every exchange between agents, keeping the workflow platform-agnostic and every interaction visible. Each agent guides you through the workflow at every step, telling you exactly which command to run, in which conversation, and what to do next.\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsdi2200262_agentic-project-management_readme_a0254acab74f.png\" alt=\"Agentic Project Management\" width=\"800\"\u002F>\n\u003C\u002Fp>\n\n## Quick Start\n\nAPM supports Claude Code, Codex CLI, Cursor, GitHub Copilot, Gemini CLI, and OpenCode.\n\nInstall the CLI:\n\n```bash\nnpm install -g agentic-pm\n```\n\nNavigate to your project directory and initialize:\n\n```bash\napm init\n```\n\nSelect your AI assistant when prompted. The CLI installs commands, guides, skills, and project artifact templates into your workspace.\n\nNext, open your AI assistant and run:\n\n```\n\u002Fapm-1-initiate-planner\n```\n\nYou can also provide context about what you want to build:\n```\n\u002Fapm-1-initiate-planner I want you to build Claude Opus 5. Make no mistakes.\n```\n\nThe Planner collaborates with you through project discovery and creates the planning documents for you to review. Once approved, it guides you to open a new conversation and run `\u002Fapm-2-initiate-manager` to begin coordinated execution. From there, each agent directs you through the workflow step by step.\n\nFor the full walkthrough, see the [Getting Started](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fgetting-started) guide.\n\n## Documentation\n\nFull documentation is available at [agentic-project-management.dev](https:\u002F\u002Fagentic-project-management.dev):\n\n- **[Introduction](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fintroduction)** - What APM is and how it works\n- **[Getting Started](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fgetting-started)** - Installation through first task cycle\n- **[Agent Types](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fagent-types)** - Planner, Manager, and Worker roles\n- **[Agent Orchestration](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fagent-orchestration)** - Communication, coordination, Memory, and Handoff mechanics\n- **[Workflow Overview](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fworkflow-overview)** - Every procedure in detail\n\nThe site also covers advanced topics like how APM's prompt and context engineering works under the hood, design principles behind the multi-agent coordination, tips and tricks for model selection and cost optimization, troubleshooting, and customization.\n\n## CLI\n\n| Command | Description |\n|---------|-------------|\n| `apm init` | Initialize with official releases |\n| `apm custom` | Install from custom repositories |\n| `apm update` | Update to latest compatible version |\n| `apm archive` | Archive current session or manage archives |\n| `apm add` \u002F `apm remove` | Add or remove assistant(s) |\n| `apm status` | Show installation state |\n\nSee the [CLI Guide](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fcli) for full details.\n\n## Customization\n\nAPM supports custom repositories for teams that want to modify the workflow. Fork the repo (for upstream sync) or use \"Use this template\" (for a clean start), adjust templates, build, release, and install with `apm custom -r owner\u002Frepo`. A [customization skill](skills\u002Fapm-customization\u002F) is included to guide AI agents through the process.\n\nSee the [Customization Guide](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fcustomization-guide) for details.\n\n### APM Auto\n\n[APM Auto](https:\u002F\u002Fgithub.com\u002Fsdi2200262\u002Fapm-auto) is an official custom adaptation of APM built for Claude Code. It replaces the user-mediated Worker model with autonomous subagent dispatch - the Manager spawns ephemeral subagents via `Agent()` to execute Tasks, reviews their output, and continues without requiring you to shuttle messages between chats. Best for prototyping, fast execution, and simpler projects.\n\n```bash\napm custom -r sdi2200262\u002Fapm-auto\n```\n\n## APM Assist\n\nThe [`apm-assist`](skills\u002Fapm-assist\u002F) skill turns your AI assistant into an APM-aware helper. Install it into any project and your assistant can explain how APM works, answer questions by reading the live documentation, detect your installation state and version, and guide migration from v0.5.x. It works with any supported platform.\n\n```bash\n# Claude Code example\nmkdir -p .claude\u002Fskills\u002Fapm-assist\ncurl -sL https:\u002F\u002Fraw.githubusercontent.com\u002Fsdi2200262\u002Fagentic-project-management\u002Fmain\u002Fskills\u002Fapm-assist\u002FSKILL.md \\\n  -o .claude\u002Fskills\u002Fapm-assist\u002FSKILL.md\n```\n\nSee the [standalone skills directory](skills\u002F) for other platforms.\n\n## Migrating from v0.5.x\n\nv1.0.0 is a ground-up redesign - the workflow, file structure, CLI, and agent roles all changed significantly. The pre-v1 codebase is preserved on the [`v0.5.x`](https:\u002F\u002Fgithub.com\u002Fsdi2200262\u002Fagentic-project-management\u002Ftree\u002Fv0.5.x) branch for reference.\n\nThe [Troubleshooting Guide](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Ftroubleshooting-guide#migrating-from-v05x) documents the recommended migration procedure. The `apm-assist` skill above can also walk you through it interactively.\n\n## Contributing\n\nContributions are welcome. See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines. Report bugs or suggest features via [GitHub Issues](https:\u002F\u002Fgithub.com\u002Fsdi2200262\u002Fagentic-project-management\u002Fissues). Reach out on Discord: `cobuter_man`.\n\n## Versioning\n\nCLI and template releases version independently but share major version for compatibility. See [VERSIONING.md](VERSIONING.md) for details.\n\n## License\n\nLicensed under the **Mozilla Public License 2.0 (MPL-2.0)**. APM is free for all uses including commercial. Improvements to core APM files must be shared back with the community. See [LICENSE](LICENSE) for full details.\n\nVersions prior to v0.4.0 were released under the MIT license. The license was updated to MPL-2.0 starting with v0.4.0.\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fsdi2200262\" target=\"_blank\">\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsdi2200262_agentic-project-management_readme_58e134115d86.png\" alt=\"CobuterMan\" width=\"150\"\u002F>\n  \u003C\u002Fa>\n\u003C\u002Fp>\n","# 智能体项目管理 (APM)\n\n[![许可证：MPL-2.0](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MPL_2.0-brightgreen.svg)](https:\u002F\u002Fopensource.org\u002Flicenses\u002FMPL-2.0) [![npm](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fv\u002Fagentic-pm)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002Fagentic-pm) [![GitHub 发布](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fv\u002Frelease\u002Fsdi2200262\u002Fagentic-project-management)](https:\u002F\u002Fgithub.com\u002Fsdi2200262\u002Fagentic-project-management\u002Freleases)\n\n*与 AI 智能体团队协作，流畅高效地管理复杂项目。*\n\n## 什么是 APM？\n\nAPM 是一个开源框架，用于借助 AI 助手管理大型软件项目。不同于在单一且日益混乱的聊天中工作，APM 将您的工作结构化为一个协调一致的系统，由不同的 AI 智能体作为一个团队共同负责规划、协调和执行。\n\n随着对话的不断推进，AI 的上下文理解能力会逐渐下降。助手可能会忘记需求、生成错误代码或凭空捏造细节。对于大型项目而言，这几乎会让持续进展变得不可能。\n\n为此，APM 协调三种专门的智能体类型，每种智能体都在独立的上下文中运行，仅处理其所需的信息：\n\n- **规划者** - 进行结构化的项目调研，并将需求分解为三份规划文档：规格书（要构建什么）、计划书（如何组织工作）以及规则（如何执行工作）。\n- **管理者** - 通过分配任务给执行者、审查已完成的工作并维护项目状态来协调执行过程。管理者基于执行摘要而非原始代码进行操作。\n- **执行者** - 在高度限定的上下文中执行特定任务。每个执行者都会收到包含所有必要信息的独立任务提示，完成任务后进行验证并汇报结果。\n\n项目状态存储在结构化的文件中，这些文件不依赖于任何智能体的上下文。当对话结束或某个智能体达到其限制时，交接机制会根据需要将当前的工作知识传递到新的实例中。这也使得已完成的 APM 会话可以被归档，并将其上下文延续到新的会话中。\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsdi2200262_agentic-project-management_readme_a0254acab74f.png\" alt=\"Agentic Project Management\" width=\"800\"\u002F>\n\u003C\u002Fp>\n\n## 快速入门\n\nAPM 支持 Claude Code、Codex CLI、Cursor、GitHub Copilot、Gemini CLI 和 OpenCode。\n\n安装 CLI：\n\n```bash\nnpm install -g agentic-pm\n```\n\n导航到您的项目目录并初始化：\n\n```bash\napm init\n```\n\n按照提示选择您的 AI 助手。CLI 会将命令、指南、技能以及项目工件模板安装到您的工作区中。\n\n接下来，打开您的 AI 助手并运行：\n\n```\n\u002Fapm-1-initiate-planner\n```\n\n您也可以提供关于您想要构建的内容的上下文：\n```\n\u002Fapm-1-initiate-planner 我希望您构建 Claude Opus 5。请务必准确无误。\n```\n\n规划者会与您协作完成项目调研，并创建规划文档供您审核。一旦批准，它会引导您开启一个新的对话，并运行 `\u002Fapm-2-initiate-manager` 以开始协调执行。从那时起，每个智能体会逐步指导您完成整个流程。\n\n有关完整教程，请参阅 [入门指南](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fgetting-started)。\n\n## 文档\n\n完整的文档可在 [agentic-project-management.dev](https:\u002F\u002Fagentic-project-management.dev) 上找到：\n\n- **[简介](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fintroduction)** - APM 是什么以及它是如何工作的\n- **[入门](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fgetting-started)** - 从安装到第一个任务周期\n- **[智能体类型](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fagent-types)** - 规划者、管理者和执行者的角色\n- **[智能体编排](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fagent-orchestration)** - 沟通、协调、记忆和交接机制\n- **[工作流概述](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fworkflow-overview)** - 每个步骤的详细说明\n\n该网站还涵盖了高级主题，例如 APM 的提示和上下文工程如何在底层运作、多智能体协调背后的设计原则、模型选择和成本优化的技巧、故障排除以及自定义等内容。\n\n## CLI\n\n| 命令 | 描述 |\n|---------|-------------|\n| `apm init` | 使用官方版本进行初始化 |\n| `apm custom` | 从自定义仓库安装 |\n| `apm update` | 更新到最新兼容版本 |\n| `apm archive` | 归档当前会话或管理归档 |\n| `apm add` \u002F `apm remove` | 添加或移除助手 |\n| `apm status` | 显示安装状态 |\n\n更多详细信息请参阅 [CLI 指南](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fcli)。\n\n## 自定义\n\nAPM 支持自定义仓库，适用于希望修改工作流程的团队。您可以 fork 该仓库以保持与上游同步，或使用“使用此模板”功能从头开始，调整模板、构建、发布，并使用 `apm custom -r owner\u002Frepo` 进行安装。其中包含一个 [自定义技能](skills\u002Fapm-customization\u002F)，可指导 AI 智能体完成整个过程。\n\n有关详细信息，请参阅 [自定义指南](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Fcustomization-guide)。\n\n### APM Auto\n\n[APM Auto](https:\u002F\u002Fgithub.com\u002Fsdi2200262\u002Fapm-auto) 是 APM 的官方自定义版本，专为 Claude Code 打造。它用自主子代理调度取代了用户介导的执行者模式——管理者会通过 `Agent()` 启动临时子代理来执行任务，审查其输出，并继续工作，无需您在不同聊天之间来回传递消息。最适合原型开发、快速执行以及较为简单的项目。\n\n```bash\napm custom -r sdi2200262\u002Fapm-auto\n```\n\n## APM Assist\n\n[`apm-assist`](skills\u002Fapm-assist\u002F) 技能可将您的 AI 助手转变为具备 APM 知识的助手。将其安装到任何项目中，您的助手便能够解释 APM 的工作原理、通过阅读实时文档回答问题、检测您的安装状态和版本，并指导您从 v0.5.x 版本迁移。它适用于任何支持的平台。\n\n```bash\n# Claude Code 示例\nmkdir -p .claude\u002Fskills\u002Fapm-assist\ncurl -sL https:\u002F\u002Fraw.githubusercontent.com\u002Fsdi2200262\u002Fagentic-project-management\u002Fmain\u002Fskills\u002Fapm-assist\u002FSKILL.md \\\n  -o .claude\u002Fskills\u002Fapm-assist\u002FSKILL.md\n```\n\n其他平台的独立技能目录请参见 [技能目录](skills\u002F)。\n\n## 从 v0.5.x 迁移\n\nv1.0.0 是一次彻底的重新设计——工作流程、文件结构、CLI 以及智能体的角色都发生了重大变化。v1.0.0 之前的代码库保留在 [`v0.5.x`](https:\u002F\u002Fgithub.com\u002Fsdi2200262\u002Fagentic-project-management\u002Ftree\u002Fv0.5.x) 分支上，供参考。\n\n[故障排除指南](https:\u002F\u002Fagentic-project-management.dev\u002Fdocs\u002Ftroubleshooting-guide#migrating-from-v05x) 记录了推荐的迁移步骤。上述的 `apm-assist` 技能也可以交互式地引导您完成迁移。\n\n## 贡献\n\n欢迎贡献。请参阅 [CONTRIBUTING.md](CONTRIBUTING.md) 以获取指南。可通过 [GitHub Issues](https:\u002F\u002Fgithub.com\u002Fsdi2200262\u002Fagentic-project-management\u002Fissues) 报告 bug 或提出功能建议。也可在 Discord 上联系：`cobuter_man`。\n\n## 版本管理\n\nCLI 和模板的发布各自独立进行版本号管理，但为了保持兼容性，它们共享主版本号。详情请参阅 [VERSIONING.md](VERSIONING.md)。\n\n## 许可证\n\n本项目采用 **Mozilla 公共许可证 2.0 (MPL-2.0)** 许可。APM 可用于所有用途，包括商业用途。对 APM 核心文件所做的改进必须与社区共享。完整详情请参阅 [LICENSE](LICENSE)。\n\nv0.4.0 之前的版本采用 MIT 许可证发布。自 v0.4.0 起，许可证更新为 MPL-2.0。\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fsdi2200262\" target=\"_blank\">\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsdi2200262_agentic-project-management_readme_58e134115d86.png\" alt=\"CobuterMan\" width=\"150\"\u002F>\n  \u003C\u002Fa>\n\u003C\u002Fp>","# Agentic Project Management (APM) 快速上手指南\n\nAgentic Project Management (APM) 是一个开源框架，旨在通过协调多个专用 AI 智能体（规划者、管理者、执行者）来高效管理复杂的软件开发项目，解决单一对话中上下文丢失和幻觉问题。\n\n## 环境准备\n\n在开始之前，请确保您的开发环境满足以下要求：\n\n*   **操作系统**：支持 macOS、Linux 或 Windows (WSL)。\n*   **Node.js**：已安装 Node.js (建议 v18 或更高版本)，因为 APM 基于 npm 分发。\n*   **AI 助手客户端**：已安装并配置好以下任一支持的 AI 编程助手：\n    *   Claude Code\n    *   Codex CLI\n    *   Cursor\n    *   GitHub Copilot\n    *   Gemini CLI\n    *   OpenCode\n\n> **注意**：目前官方未提供特定的中国镜像源，请使用标准的 npm 源。如果网络受限，可配置国内 npm 镜像（如淘宝镜像）加速安装：\n> ```bash\n> npm config set registry https:\u002F\u002Fregistry.npmmirror.com\n> ```\n\n## 安装步骤\n\n1.  **全局安装 APM CLI**\n    打开终端，运行以下命令安装命令行工具：\n    ```bash\n    npm install -g agentic-pm\n    ```\n\n2.  **初始化项目**\n    进入您的目标项目目录，并运行初始化命令：\n    ```bash\n    apm init\n    ```\n\n3.  **配置 AI 助手**\n    运行上述命令后，CLI 会提示您选择正在使用的 AI 助手。根据提示完成选择后，工具会自动将必要的命令、指南、技能文件及项目模板注入到您的工作区中。\n\n## 基本使用\n\n安装完成后，您需要在 AI 助手的对话界面中通过特定指令启动工作流。\n\n1.  **启动规划阶段 (Planner)**\n    打开您的 AI 助手（如 Cursor 或 Claude Code 的聊天窗口），输入以下指令启动规划者智能体：\n    ```\n    \u002Fapm-1-initiate-planner\n    ```\n    \n    您也可以直接附带项目需求描述：\n    ```\n    \u002Fapm-1-initiate-planner I want you to build a simple todo app with React.\n    ```\n\n2.  **审查与确认**\n    Planner 智能体将与您协作进行项目探索，并生成三份核心文档供您审查：\n    *   **Spec** (构建什么)\n    *   **Plan** (如何组织工作)\n    *   **Rules** (如何执行工作)\n    \n    仔细阅读并确认这些文档无误。\n\n3.  **启动执行阶段 (Manager)**\n    规划文档批准后，Planner 会指导您开启一个新的对话窗口。在新窗口中运行以下指令以启动管理者智能体，开始协调任务执行：\n    ```\n    \u002Fapm-2-initiate-manager\n    ```\n\n4.  **跟随工作流**\n    此后，AI 智能体会逐步引导您完成剩余流程。Manager 会将具体任务分配给 Worker 智能体，您只需按照提示在不同对话间切换或确认即可，无需手动传递大量上下文信息。\n\n> 更多详细的工作流说明和高级配置，请访问官方文档：[agentic-project-management.dev](https:\u002F\u002Fagentic-project-management.dev)","某初创团队的技术负责人正带领两名开发者，试图在两周内从零构建一个包含用户认证、实时数据看板及支付网关集成的高复杂度 SaaS 平台。\n\n### 没有 agentic-project-management 时\n- **上下文严重丢失**：随着对话轮数增加，AI 逐渐遗忘早期的核心业务规则，导致生成的代码与初始需求背道而驰，甚至出现幻觉功能。\n- **协作混乱无序**：所有规划、编码和审查工作挤在同一个聊天窗口中，AI 难以区分当前是处于“设计阶段”还是“修复 Bug 阶段”，输出内容杂乱无章。\n- **状态维护困难**：项目进度依赖人工记忆或分散的笔记，一旦会话中断或切换助手，之前的逻辑推导过程无法有效继承，不得不反复重述背景。\n- **代码质量失控**：由于缺乏独立的审查环节，AI 直接生成的代码往往包含隐蔽错误，开发者需花费大量时间手动排查而非专注于架构设计。\n\n### 使用 agentic-project-management 后\n- **结构化知识沉淀**：Planner 代理将需求拆解为独立的规范（Spec）、计划（Plan）和规则（Rules）文档，确保核心逻辑始终清晰且可追溯，不受对话长度限制。\n- **角色分工明确**：Manager 代理专职协调任务分配与进度，Worker 代理仅在狭义上下文中执行具体编码，三者各司其职，彻底消除了流程混淆。\n- **无缝会话接力**：项目状态存储在外部文件中，通过 Handoff 机制，新开启的会话能瞬间继承完整的项目认知，支持多人多轮次高效协作。\n- **自动化质量闭环**：Worker 完成任务后自动验证并汇报，Manager 基于执行摘要进行审核而非原始代码流，显著提升了交付物的可靠性与一致性。\n\nagentic-project-management 通过将混沌的单聊转化为精密的多代理协作流水线，让大型软件项目的 AI 辅助开发变得可控、可持续且高质量。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fsdi2200262_agentic-project-management_b0a97be3.png","sdi2200262","CobuterMan","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fsdi2200262_0a46bcee.jpg","The name is Man... CobuterMan.",null,"Leoforos Alexandras 160","https:\u002F\u002Fgithub.com\u002Fsdi2200262",[81],{"name":82,"color":83,"percentage":84},"JavaScript","#f1e05a",100,2192,208,"2026-04-17T18:15:47","NOASSERTION","未说明",{"notes":91,"python":89,"dependencies":92},"该工具是一个基于 CLI 的框架，用于协调现有的 AI 编码助手（如 Claude Code, Cursor, GitHub Copilot 等），而非独立的本地大模型运行环境。因此无需特定 GPU 或 Python 环境。安装需要 Node.js 环境以运行 npm 命令。用户需自行拥有并配置所选择的 AI 助手服务账号。",[93,94],"Node.js (需支持 npm)","agentic-pm (CLI)",[13,14],"2026-03-27T02:49:30.150509","2026-04-20T16:21:34.543917",[],[100,105,110,115,120,125,130,135,140,145,150,155,160,165,170,175,180,185,190,195],{"id":101,"version":102,"summary_zh":103,"released_at":104},323794,"v1.0.0","v1.0.0 对 APM 工作流进行了全面重新设计。完整规范请参阅[文档](https:\u002F\u002Fagentic-project-management.dev)。\n\n### 重大变更\n\n* **工作流重新设计。** 两阶段工作流：规划阶段（上下文收集 + 工作分解）生成三份规划文档（规格书、计划书、规则）。实施阶段则循环进行任务分配、任务执行、任务记录和任务评审，并支持向多个 Worker 批量及并行派发任务。\n* **代理角色调整。** 设置代理 → 规划者。实施代理 → 工人。移除临时代理。管理者保持不变。\n* **新增 9 条命令**（原为 5 条）：apm-1 至 apm-9。\n* **全新工件结构。** 包括规格书、计划书、规则、跟踪器和记忆层级，取代原有的实施计划和记忆根。\n* **平台支持扩展** —— Claude Code、Codex CLI、Cursor、GitHub Copilot、Gemini CLI 和 OpenCode。\n* **CLI 重新设计。** 新增命令：archive、add、remove、status、custom。实现按文件粒度的安装追踪。\n* **版本解耦。** CLI 与模板独立版本化。\n\n### 新增功能\n\n* **新增 Codex CLI 支持** 作为新受支持的平台\n* 基于文件的代理间通信总线\n* 会话归档与续接功能\n* 上下文重建的恢复命令\n* 结构化上下文传递的手动交接机制\n* 独立技能模块（apm-assist、apm-customization）\n* 自定义仓库支持\n* 带占位符系统的构建流水线\n* 模块化 CLI 架构\n\n### 移除内容\n\n* 临时代理及委托相关命令\n* 实施计划、记忆根、指南\u002F格式文件\n* 取消对 Windsurf、Kilo Code、Roo Code、Auggie CLI、Google Antigravity 和 Qwen Code 的支持\n* 废弃旧版构建系统","2026-04-09T09:47:13",{"id":106,"version":107,"summary_zh":108,"released_at":109},323795,"v1.0.0-test-23","v1.0.0 稳定版之前的最终预发布版本。对所有运行时模板进行了清理、标准化合规性检查以及行为上的优化。\n\n### 模板变更\n\n- **先调查后修复的调试流程** — Worker 在尝试修复之前会先调查根本原因。每次迭代仅执行一次有针对性的修复。若修复失败，Worker 会启动一个带有结构化指令的调试子代理（包括错误输出、已调查内容、预期与实际差异，以及追踪根本原因的指示）。Worker 会在应用子代理的发现结果前对其进行验证。修复循环现已重构为带编号的流程步骤。\n\n- **子代理发现结果的验证** — 三种角色现在都会在采取行动前阅读关键引用文件来验证子代理的发现，这一模式与 Planner 在上下文收集阶段的现有做法保持一致。\n\n- **阶段验证重新设计** — 验证工作已从工作分解中的计划任务转移至 Manager 的运行时职责。Planner 会在计划备注中记录观察结果；Manager 则在阶段结束时决定是否需要进行验证。任务评审新增 §2.8：重新执行关键的 Worker 检查、测试未覆盖的边界情况、进行全面的端到端检查，并阅读源文件以确认代码库状态。\n\n- **自然语言条件语句** — 所有运行时文件及 WRITING.md 中的箭头符号已被自然语言中的逗号替代。决策规则标签和评审结果标签也已替换为条件性描述性语句。\n\n- **上下文范围界定与 Manager 初始化** — Worker 的操作规则不再指定具体文档，而是改为笼统的“规划或协调文档”。Manager 通过 Handoff Bus（此前由 Tracker 模板状态判断）来区分首次实例与后续实例，这与 Worker 的处理方式保持一致。\n\n- **计划分析与 Planner 备注** — 每个阶段包含自成一体的模块，并明确交付物与任务的对应关系。计划评审环节已被移除。Spec 备注聚焦于项目环境，Plan 备注则关注工作结构方面的观察，以 Manager 的认知视角呈现。\n\n- **基准代码质量默认设置** — 任务执行部分新增一段引言：编写整洁易维护的代码，遵循语言和框架的最佳实践，采用具有描述性的命名方式，并逐步构建功能。若另有明确规定，则相关内容将移至任务提示和规则中。\n\n- **规则放宽与恢复机制** — 规则放宽至“所有或大多数任务”，并辅以条件性表述。恢复指令的措辞得到强化，明确列出所需文档清单。触发条件包括对话记录被清空等情形。\n\n- **标准审计合规性** — 实施两轮审计。修正内容包括：双冒号规则、语态一致性、去重处理（删除了重复阐述 §2 内容的 §5.1）、术语同义词、过时的交叉引用，以及 WORKFLOW 对齐方面的差距。阶段完成环节已整合进持续协调流程中。\n","2026-04-08T20:00:15",{"id":111,"version":112,"summary_zh":113,"released_at":114},323796,"v1.0.0-test-22","最后一条，保证是最后一次：里程碑验证关卡、失败点清理、计划分析的递进推导。\n\n### 模板变更\n\n- **里程碑验证关卡** — 以逐需求验证为主要机制。里程碑评审关卡（在各阶段边界进行的整体性验证）则是另一项独立的关注点：规划师会在必要时将其作为问题呈现出来，但仅当用户明确批准时才会将其纳入计划中。此类验证行为不属于规则范畴（仅限于任务范围，而非全局通用）。\n\n- **移除 `failure_point` 字段** — 该字段在术语体系和任务日志中均为冗余内容，从未在任何测试会话中承载过有效数据。部分完成或失败的结果已在任务日志的文本描述部分予以说明。\n\n- **计划分析的递进推导** — 现在每个分析章节都以前一章节为基础逐步推导：领域信息用于界定阶段边界，阶段目标则进一步分解为具体任务。阶段分析从已识别的领域出发，结合其特性推导出自然的阶段性推进点。任务分析则基于阶段目标所分解出的任务展开深入细化，而非直接采用既定的任务列表。\n\n- **指南与步骤的区分** — 在计划格式中进一步强化：步骤是指按顺序执行的操作，而指南则提供背景信息和约束条件。两者应保持分离。\n\n### CLI 变更\n\n- 在 `--help` 输出及安装后提示信息中引用了 `apm-assist` 技能（尚未发布至 npm）。","2026-04-04T15:32:33",{"id":116,"version":117,"summary_zh":118,"released_at":119},323797,"v1.0.0-test-21","希望这是最后一份：单遍工作分解分析、权限澄清、子代理对齐调试、代理解析用户体验，以及跨平台格式化。\n\n### 模板变更\n\n- **工作分解单遍分析** — 每个文档部分按照分析 → 读取目标文件 → 写入的流程执行。规范\u002F计划头区块已合并到写入步骤中。计划评审被明确为写入后的验证环节。\n  \n- **依赖分析自然描述** — 推理过程着重描述关系，而非图形渲染属性。图形格式化在写入步骤中完成。\n\n- **任务指导引用规范** — 关于共享设计决策的指导直接引用规范章节，不再重复说明，仅在必要时补充领域特定上下文。这一原则已在计划标准和各任务推理框架中进一步强化，并新增了“重复说明”常见错误提示。\n\n- **计划备注中的调度相关性** — 指导内容恢复为突出与调度相关的观察点：独立任务、顺序链、汇聚点以及相对复杂度。\n\n- **无提示路径上的权限问题** — 发现声称对其他文档具有权限的材料，需像其他发现的材料一样，由用户确认。\n\n- **任务提示中的子代理调试** — 任务迭代字段与修正循环保持一致：子代理作为第二步，拥有完整上下文；针对独立根本原因，则并行启动多个子代理。\n\n- **后续深化处理** — 后续任务利用管理器更广泛的上下文，而非重复原任务内容。\n\n- **工作者验证阶段转换** — 工作者在从执行阶段转入验证阶段时会进行沟通。\n\n- **工作者注册简化** — 用户端输出中移除了总线路径和目录列表。\n\n- **代理ID解析** — 匹配规则更加宽松：优先精确匹配，其次为前缀匹配，最后选择最合理的可能匹配。严格限制最小字符数的要求已被取消。\n\n- **代码块中的命令** — 可操作命令使用代码块以方便复制；引用类内容则采用内联代码格式。\n\n- **其他** — `\u002Fapm-4` 准备就绪信号明确为无参数形式；修复了 `\u002Fapm-8` 代码库状态比较范围的问题；同时明确了 `\u002Fapm-8` 命令需在完成管理器对话后方可使用；此外，还新增了关于规则引用不可访问文档的常见错误提示。\n\n### 仓库变更\n\n- 将 apm-assist 技能与 llms.txt 入口点整合，并合并迁移技能。\n- 在支持平台文档中添加 Codex CLI 相关内容，并恢复 js-yaml 依赖项。","2026-04-03T18:26:27",{"id":121,"version":122,"summary_zh":123,"released_at":124},323798,"v1.0.0-test-20","Worker 初始化重新设计、总线协议可靠性、调试委托、工作树隔离、交接连续性、规划器分解推理，以及 Codex CLI 平台支持。\n\n### 模板变更\n\n- **Worker 初始化时在注册时检查总线状态** — 在身份解析完成后，Worker 会同时检查任务总线和交接总线，以确定初始化路径。如果已有任务提示正在等待，则立即开始执行，无需再调用 `\u002Fapm-4-check-tasks`。同样通过总线检查即可检测到任务中或批次中的交接情况，并由交接提示描述每个任务的完成状态。这不仅消除了每次首次调度时的一次往返通信，还修复了批次中间交接时存在的漏洞：此前新到达的 Worker 必须重新读取并手动跳过已完成的任务。\n\n- **程序节点处的“先读后写”总线规则** — “先读后写”的总线协议规则虽然已纳入通信技能，但在任务分配和任务日志记录的实际写入步骤中并未得到强化。代理经常违反该规则，仅靠平台工具的强制约束才能被发现。现在，每个总线写入步骤都会明确地先进行读取操作。同时，总线生命周期也得到了进一步澄清：目录应在规划阶段初始化，后续代理不再重复创建。\n\n- **将调试子代理作为第二步，而非最后手段** — 当 Worker 的首次修复尝试未能解决问题时，下一步是启动一个携带完整累积上下文的调试子代理，而不是继续在主上下文中反复迭代。测试中曾出现过因调试循环过于冗长而占用大量上下文资源的情况。当存在多个独立的根本原因时，可分别启动不同的子代理并行排查。\n\n- **工作空间发现顺序** — 原本关于在阅读上下文收集指南之前禁止深入代码库探索的规定，被夹杂在句子中间。现将其提升为一条独立指令：轻量级工作空间扫描期间，不得读取源代码，也不得分派子代理。\n\n- **工作树标准进一步收紧** — 每个并行调度单元都将拥有自己的工作树（从而避免出现某些 Worker 仍留在主目录中的不对称方案）。针对工作树调度，明确了 `.apm\u002F` 路径的解析方式：Worker 在工作树中执行代码相关操作，但所有 `.apm\u002F` 路径的解析均基于项目根目录进行。即使 `.apm\u002F` 已被 Git 跟踪，运行时操作的目标仍然指向项目根目录，以便 Manager 能够在合并前查看日志。此外，工作树清理的顺序及其理由也被明确列出。\n\n- **交接产物的区分** — 关于如何继续工作的指导信息已从交接日志（过去式记录）转移到交接提示（可操作的状态）。先前阶段的日志报告现在仅在确实存在先前阶段的情况下才会生成，从而避免在第一阶段出现无意义的表述。Manager 对交接的检测机制也进行了更新，当不适用“上一阶段备注”时，不再强制要求提供该信息。\n\n- **任务指导引用规范** — 规划任务的指导内容可以直接引用规范中的相关章节，以说明共享的设计决策，而无需重复阐述。Manager 将在构建任务提示时解析这些引用。此举减少了……之间的重复。","2026-04-02T18:51:21",{"id":126,"version":127,"summary_zh":128,"released_at":129},323799,"v1.0.0-test-19","渐进式工作流：上下文感知、自然沟通重构、工作空间发现与重组，以及跨平台总线兼容性。\n\n### 模板变更\n\n- **工作流上下文感知** — 现在“上下文收集”和“工作分解”指南均新增了§2.1 工作流上下文章节，帮助规划者逐步理解后续步骤。在上下文收集阶段，规划者需明确所收集的上下文将用于生成面向不同受众的三份文档，而协调工作则属于管理者在运行时的职责范围——因此提问应聚焦于内容本身，而非协调机制。而在工作分解阶段，这种认知进一步深化，具体体现在各文档的使用方式上：管理者会按任务提取规范中的内容生成任务提示，并在运行时对其进行丰富；而工作者则专注于其任务提示及规则说明。\n\n- **工作空间发现移至规划者启动环节** — 轻量级工作空间扫描、归档检查以及文档权威性评估，现均在规划者的启动命令（apm-1 §2）中执行，即在阅读“上下文收集”指南之前完成。规划者首先确定哪些材料具有权威性——这些权威性来源可以是预先准备好的上下文信息，也可以由用户确认——随后再结合权威文档一起阅读上下文收集指南。此举可避免因文件边界问题而导致权威性判断丢失。\n\n- **自然沟通与透明推理** — 代理问候语由枚举职责清单改为核心角色描述，更加自然流畅。透明推理则从“自言自语式思考”调整为向用户呈现分析结果，供其理解、复核与审计。所有面向用户的输出中，均禁止出现指南章节标题、流程名称及框架术语。同时，所有六份指南中的流程内省部分已实现标准化。\n\n- **规划者观察记录传递** — 规划者可将其客观观察（如版本控制模式、协调偏好等）以引用块的形式记录在规范的工作空间头部和计划的依赖图头部。这些观察仅供管理者参考，并非必须采纳的惯例。\n\n- **任务提示丰富策略** — 规范提取标准进行了重组：嵌入来自规划文档及权威来源的内容（这些内容不应由工作者自行发现），并提供针对代码库内容的引用及精准阅读指导（由工作者直接查阅相关文件）。\n\n- **工作笔记与记忆笔记提炼** — 阶段结束时采用三重分流模式：工作笔记可汇入阶段总结性文字（作为历史记录）或记忆笔记（用于记录具有长期协调影响的观察）。阶段总结旨在记录已发生的事情；而记忆笔记则聚焦于对未来有重要意义的内容。\n\n- **非APM代理参与** — 针对独立代理加入会话的情况，简化了总线集成指南。管理者能够检测到意外报告（任务评审§2.8），并通过普通任务分配指令（任务分配§2.7）指派后续工作。非APM代理无需进行日志记录、交接操作，也不参与工作者跟踪流程。\n\n- **管理者监控与完成情况** — 管理者负责监控工作","2026-03-30T19:18:26",{"id":131,"version":132,"summary_zh":133,"released_at":134},323800,"v1.0.0-test-18","版本控制所有权重新设计、文档权限流程、子代理验证以及 CLI 安全加固。\n\n### 模板变更\n\n- **版本控制从规划者移至管理者** — 规划者不再询问版本控制相关问题或向追踪器写入信息。它仅将关于版本控制的事实性观察以备注形式记录在规格说明中。所有版本控制的初始化工作均由管理者在首次初始化时负责：检查 Git 状态，在一次审批流程中连同理解摘要一并展示约定规范，并在获得批准后写入规则和追踪器。\n\n- **文档权限流程** — 规划者可在初步了解阶段自由阅读发现的材料。若需深入探索（派遣子代理），则必须获得发起上下文的授权或用户确认。\n\n- **子代理验证周期** — 规划者会直接读取引用文件以核实子代理的声明。若验证结果与现有发现相矛盾，则会派遣并依次验证相应的子代理。\n\n- **实施阶段意识增强** — 规划者的初步了解阶段涵盖多名工作者、任务级验证及分支模型，以避免在情境收集过程中提出误导性问题。\n\n- **工作者跟踪名称变更** — 整个系统中将“代理跟踪”更名为“工作者跟踪”，因为本追踪器部分仅跟踪工作者。\n\n- **计划评审重构** — 写作完成后进行结构、工作量及一致性检查。发现问题后先予修正，再提交给用户。\n\n- **移除任务分配中的版本控制验证部分** — 该部分与管理者的初始化工作重复。各章节重新编号，并更新交叉引用。\n\n- **按标准审计** — 对运行时文件中的任务生命周期术语、破折号、双冒号、交叉引用及间距等进行了统一修正。\n\n### CLI 变更\n\n以下 CLI 变更已包含在 `agentic-pm@1.0.0-test-3` 中。这些变更属于可选内容——使用现有的 `v1.0.0-test-2` CLI 也能正常工作。如需更新，请执行：\n\n```bash\nnpm install -g agentic-pm@1.0.0-test-3\n```\n\n- **路径遍历防护** — 如果 ZIP 文件条目解析到项目目录之外，系统将发出警告并阻止该操作。清单模式会拒绝在 `configDir` 和 `bundle` 字段中出现的路径遍历序列。\n\n- **`apm add` 的安全免责声明** — 自定义安装现在会在执行 `apm add` 时显示免责声明，与 `apm custom` 和 `apm update` 的行为保持一致。\n\n### 仓库变更\n\n- SECURITY.md：修正了路径遍历行为，并新增了已知限制。\n- README.md：介绍了分叉与模板选项及其各自的优缺点。\n- VERSIONING.md：修正了自定义更新的行为。\n- 更新了迁移与自定义技能相关内容。\n- 移除了过时的 ISSUES.md 标准文件。","2026-03-27T11:30:51",{"id":136,"version":137,"summary_zh":138,"released_at":139},323801,"v1.0.0-test-17","针对规划器的 VC 处理、文档权限以及推理框架呈现的小幅调整。\n\n### 模板变更\n\n- **VC 始终分支** — 工作者始终在功能分支上操作，绝不直接在主分支上进行更改。无论采用何种调度模式，分支使用都是默认行为；规划器仅会询问相关约定。实施阶段的表述已更新为正面引导式，以避免禁止列表泄露到规则中。\n\n- **文档权限与验证** — 对于发现的文档，在开始探索之前需获得用户的确认。子代理的验证及向用户展示的内容已直接嵌入到流程步骤中。\n\n- **推理框架** — 在 WRITING.md 中，沟通技能明确了哪些是规定的标题，哪些是需要涵盖的方面。当指定了标题时，代理会遵循既定结构；而当仅描述了需要涵盖的方面时，则可自由选择呈现形式。\n\n- **规划器的认知** — 添加了 .apm\u002F 脚手架上下文。强化了任务分析的覆盖要求。","2026-03-25T21:17:42",{"id":141,"version":142,"summary_zh":143,"released_at":144},323802,"v1.0.0-test-16","验证类型移除、版本控制重构、沟通方式全面升级，以及提升 Planner 的意识与能力。\n\n### 模板变更\n\n- **移除验证类型分类** — 整个工作流中不再区分三种验证类型（程序化、工件、用户）。验证标准现为具体的、针对任务的检查项，不再带有类别标签。行为顺序保持不变：自主检查将在涉及用户之前完成。“用户评审”扩展为“用户参与”，涵盖判断（设计审批）和操作（执行外部检查、确认平台行为）两个方面。工作人员在暂停时需明确告知所需内容、预期结果及反馈事项。\n\n- **版本控制重构** — 提交规范移至《规则》文档，分支命名规范移至《追踪器》文档。版本控制表格现按仓库划分（每个仓库一行），支持多仓库工作空间。消除了任务追踪与版本控制表格之间基础分支的重复记录。工作人员仅在分配的分支上操作，不得自行创建分支、管理工作树、推送或合并代码。\n\n- **沟通方式全面升级** — 移除行动指令的区块引用格式。代理间以自然语言交流，命令和数值部分采用代码格式。审批节点会引导用户了解下一步流程。推理框架的呈现方式改为描述行为，而非使用命名类别。\n\n- **Planner 的意识与权限** — 新增消费者模型：Planner 明确 Worker 仅可见任务和规则，任务分配由 Manager 制定，协调决策在运行时进行。子代理的发现结果在纳入前必须先与代码库核对，并向用户汇报。当存在多种可行方案时，Planner 将提供备选方案并给出建议，而非自主选择。工作空间扫描过程中发现的文档，在深入分析前需先征得用户确认。\n\n- **流程与工件结构** — 流程章节在包含明确的节点、循环或条件分支时，将增加简要的结构说明。Spec 和 Plan 中的横向规则仅保留一个标题与内容的分隔符。工件模板简化为仅保留 YAML 和标题。Tracker 的作用范围限制了 Planner 无法填充属于 Manager 的部分。会话总结结果由阶段完成数量机械生成。\n\n- **按标准的审计修复** — 对运行时文件中的术语大小写、双冒号违规、语态修正、引用格式及去重等问题进行了统一修复。\n\n### 标准变更\n\n- **WORKFLOW.md** — 从任务分解中移除验证类型；执行流程采用行为顺序；新增多仓库版本控制例外情况。\n- **TERMINOLOGY.md** — 删除“验证方法”这一概念定义。\n- **STRUCTURE.md** — 允许在具有承重意义时使用流程概览；将重新编号规则与 §5 特有的提升要求分离。\n- **WRITING.md** — 更新了推理框架的描述方式；箭头标签标点符号调整。","2026-03-23T20:39:16",{"id":146,"version":147,"summary_zh":148,"released_at":149},323803,"v1.0.0-test-15","> **重新发布说明：** 原始 v1.0.0-test-15 版本（2026年3月22日）是在提交 `6046e15` 时发布的，当时尚未推送下文所述的模板变更。发布说明中描述了预期范围，但实际发布的版本仅包含工作空间评估相关的提交（`6536037`）。本次重新发布的目标提交为 `8812662`，该提交包含了发布说明中所描述的所有变更。说明内容未作修改，仅更正了目标提交。\n\n---\n\n模板套件简化、新增规划功能，并通过全面的标准合规性检查。\n\n### 模板变更\n\n- **模板套件精简** — 指南、命令和技能中的冗余内容已被移除：导航章节、与小节标题重复的流程概要、自文档化的 YAML 字段注释，以及重述操作标准的内容指南。运行时文件更加精简，同时保留了所有操作指导。\n\n- **按阶段重构 WORKFLOW.md** — 将原本扁平化的 §6 流程拆分为 §6 规划阶段、§7 实施阶段和 §8 移交与持续性，并在各流程中加入内联的 `**运行时：**` 注释，以链接每个流程与其对应的实施文件。子代理使用部分已合并至 §1 设计阶段。实施阶段的概述则被压缩，以符合其作为高层次摘要的角色。\n\n- **工作空间评估与概览** — 在情境收集环节新增了一项流程，用于在问答轮次开始前正式扫描工作空间：包括目录结构、Git 仓库、现有材料、`{RULES_FILE}` 的发现与讨论，以及 `.apm\u002F` gitignore 配置。评估结果将输入规范中的全新工作空间概览部分，使管理者无需亲自探索即可全面了解项目环境。\n\n- **更丰富的任务分解逻辑** — 任务识别指南将阶段目标与任务边界（交付成果、人员映射、独立性评估、分解启发式方法）紧密衔接。阶段顺序推理可在计划出现缺陷之前及时发现排序错误。完整性段落确保所有收集到的情境信息均通过明确的映射关系分配到三个产出物中。边界划分推理要求规划者在可视化分析过程中，根据信息的最终归属（规范、任务指南或规则）对每一条情境信息进行分类。计划评审现新增工作量细分与依赖关系审计的推理框架、计划摘要步骤，以及解释依赖图如何支持管理者决策的派单模式注释。\n\n- **迭代与协作的清晰化** — 任务执行部分明确了工作者应在何时暂停（用户验证、外部行动、迭代停滞）以及何时可自主继续。纠错循环遵循 WORKFLOW 序列：先由子代理处理，若问题仍未解决再升级至用户。规则更新的时间安排也已对齐：工作者首先记录并上报，然后在回合结束时提出规则补充建议。自动压缩恢复机制现在会从管理者处获取更为全面的依赖关系上下文。\n\n- **计划中将“代理”更名为“工作者”** — 计划头部、交叉-","2026-03-22T13:50:05",{"id":151,"version":152,"summary_zh":153,"released_at":154},323804,"v1.0.0-test-14","Version control formalization, Message Bus terminology, APM_RULES rename, user-facing docs overhaul, and standalone skills.\r\n\r\n## Template Changes\r\n\r\n- **Version control formalized throughout the workflow** — the Planner now establishes VC during the Planning Phase: detects git state during codebase exploration, captures commit and branch conventions in Rules (proposing `type: description` commits and `type\u002Fshort-description` branches as defaults), initializes the repository, and records VC state in the Tracker. The Manager verifies and coordinates (branches, merges, cleanup) but no longer discovers conventions. Workers commit following Rules conventions but do not push, create branches, or merge. Manager commits on behalf for successful tasks with missed commits. Multi-repository workspaces supported with per-repo conventions. Users can decline VC (sequential dispatch only) or enable it mid-session.\r\n- **Message Bus elevated as formal term** — \"bus system\" renamed to \"Message Bus\" across all templates and standards. Added to TERMINOLOGY.md as a formal defined term.\r\n- **APM_STANDARDS renamed to APM_RULES** — namespace block in Rules file renamed throughout all templates and standards for consistency with documentation terminology.\r\n- **Four-procedure task cycle** — WORKFLOW.md corrected to specify four procedures (Task Assignment, Task Execution, Task Logging, Task Review) matching TERMINOLOGY.md and the guide structure.\r\n\r\n## Repository Changes\r\n\r\n- **Standalone skills directory** (`skills\u002F`) — two independently installable skills added:\r\n  - `apm-migration` — guides AI agents through migrating from v0.5.x to v1, with archive-based context carryover and revert guidance\r\n  - `apm-customization` — guides AI agents through customizing APM templates in forked repositories\r\n- **Documentation moved to dedicated repo** — docs now live in [apm-website](https:\u002F\u002Fgithub.com\u002Fsdi2200262\u002Fapm-website\u002Ftree\u002Fmain\u002Fdocs), fully updated for v1 including VC formalization\r\n- **README, CONTRIBUTING, CHANGELOG, VERSIONING rewritten** — all user-facing docs aligned with v1 stable release, new social card banner\r\n- **Issues #85 and #45 closed** — platform reduction shipped, v0.5.x feedback superseded by v1","2026-03-20T19:18:53",{"id":156,"version":157,"summary_zh":158,"released_at":159},323805,"v1.0.0-test-13","UX improvements for action directives, context gathering, and Rules integration.\n\n## Changes\n\n- **Action directives** — new blockquote format for directing users to take action, both APM workflow navigation and general actions. Platform-specific new-chat guidance via `{NEW_CHAT_GUIDANCE}` placeholder (Claude Code, Copilot, Gemini CLI, Cursor, OpenCode).\n- **Open-ended elicitation in context gathering** — each question round's final exchange includes a dynamically generated catch-all question to surface context the targeted questions missed. Folded into existing exchanges, never standalone.\n- **User corrections → Rules pipeline** — Workers note user corrections as important findings, propose Rules addition at end of turn after logging and reporting. Non-blocking: user can ignore and proceed.\n- **Instruction accuracy in Task Prompts** — new section clarifying that objectives are authoritative but detailed steps may contain inaccuracies; Workers validate against actual codebase state.\n- **Stage directory creation simplified** — Workers write directly to `log_path`; file write creates parent directories. No explicit mkdir step.\n- **Project completion signal** — Manager sets `status: complete` in Tracker YAML frontmatter; apm-8 uses it as authoritative outcome signal.\n- **Session summary timestamp** — `date` field upgraded to full ISO 8601 datetime.","2026-03-16T11:23:18",{"id":161,"version":162,"summary_zh":163,"released_at":164},323806,"v1.0.0-test-12","Agent-guide separation of concerns and subagent guidance improvements.\n\n## Changes\n\n- **Guide scoping enforced** — each agent reads only its own guides. Cross-agent guide references removed. Init commands enforce context scope with an explicit rule.\n- **VC skill dissolved into guides** — Manager VC operations to task-assignment\u002Ftask-review, Worker VC standards to task-execution. Commit terminology rule: no APM vocabulary in commits, branches, or source comments.\n- **Communication skill stripped to protocol** — agent-specific procedures (delivery, reporting, bus init, handoff bus) migrated to respective guides. Skill retains only bus identity, formats, and shared protocol.\n- **Subagent guidance strengthened** — placeholders rewritten to emphasize spawning dedicated instances with own context windows rather than inline searching.","2026-03-14T23:16:03",{"id":166,"version":167,"summary_zh":168,"released_at":169},323807,"v1.0.0-test-11","Major update from `v1.0.0-test-10` — new session continuation workflow, full CLI redesign, recovery command, comprehensive standards compliance, and platform decision corrections across all 5 targets.\n\n## Installation\n\n**CLI:**\n```bash\nnpm install -g agentic-pm@1.0.0-test-2\n```\n\n**Templates (via the new CLI):**\n```bash\napm init -t v1.0.0-test-11\n```\n\nThe CLI was redesigned alongside these templates. If you're on an older CLI version, update to `v1.0.0-test-2` first — the new template commands reference CLI features (`apm archive`, `apm add`) that don't exist in earlier versions.\n\n## Templates\n\n### New: Session Continuation\n\nCloses the APM session loop — when a project finishes (or at any point), you can summarize and archive it, then start a follow-up session with access to previous context.\n\n- **`\u002Fapm-8-summarize-session`** — new command that spawns a non-APM agent to explore the current `.apm\u002F` directory, source code, and planning artifacts, then produces a session summary document. Not a formal APM role — just an ad-hoc assistant to close the loop.\n- **`apm-archive-explorer`** — new agent file that Planners spawn as a subagent to explore archived sessions and extract relevant context without loading full archives into the context window.\n- **Archive detection in Planner context gathering** — Planners check `.apm\u002Farchives\u002F` before starting question rounds and offer to explore relevant previous sessions.\n- **Manager completion flow** — Managers recommend running `\u002Fapm-8-summarize-session` in their completion message.\n\n### New: Recovery from Auto-Compaction\n\n- **`\u002Fapm-9-recover`** — new troubleshooting command that reconstructs agent context after the platform compacts a conversation mid-session. Supports flexible role detection (arguments, context clues, or ask User). Replaces a non-functional Framework Rule.\n\n### Agent Instance Terminology\n\n- \"Agent sessions\" renamed to \"agent instances\" (numbered lifetimes within a single APM session). \"Session\" is now reserved for the overall APM project. \"Chat\" replaces \"session\" where referring to platform conversations.\n\n### Standards Compliance and Workflow Optimization\n\n- **Voice\u002Fperson distinction** codified in WRITING.md: standards use 3rd-person, runtime files use 2nd-person addressing the reading agent. Applied across all commands and guides.\n- **Per-standard audit** (45 findings): em-dashes to hyphens, \"follow\" to \"per\", Stage\u002FMemory capitalization, section cross-reference titles, batch constraint softening, stop condition additions.\n- **Memory term elevation**: Memory is now a formal term in TERMINOLOGY.md, scoped to `.apm\u002Fmemory\u002F` contents.\n- **Context gathering**: prohibited planning vocabulary during information-gathering rounds, round depth minimum, subagent verification as procedural step.\n- **Work breakdown**: renamed to Per-Stage Task Analysis, Plan Header Format spec added, User involvement in Task Scope reasoning.\n- **Task execution**: natural subagent framing replacing mechanical thresholds.\n- **Task review**: Memory Note filtering (ephemeral vs durable), last-Task shortcut for Stage summary.\n- **Placeholder rename**: `{AGENTS_FILE}` renamed to `{RULES_FILE}`.\n\n## Build Pipeline\n\n- **Agents directory** support added for all 5 targets (flat files; Copilot uses `.agent.md` extension).\n- **New placeholders**: `{AGENT_PATH}`, `{AGENTS_DIR}`, `{GUIDES_DIR}`, `{ARCHIVE_EXPLORER_GUIDANCE}`, `{SUBAGENT_GUIDANCE}`.\n- **Platform decision fixes** across all 5 targets:\n  - Claude Code: subagent tool `Task` → `Agent`\n  - Gemini CLI: `{RULES_FILE}` → `GEMINI.md`, subagent syntax → `@codebase_investigator`, config note → `experimental.enableAgents`\n  - GitHub Copilot: `{ARGS}` → `${input:args}`, agent files use `.agent.md` extension\n  - OpenCode: commands directory → plural `.opencode\u002Fcommands`, subagent tool → capitalized `Task`\n- **`postInstallNote`** field in release manifest — CLI displays per-platform warnings after install (e.g., Gemini gets a subagent config reminder).\n\n## CLI (`v1.0.0-test-2`)\n\nFull restructure of the CLI. The command surface, installation flow, and output have all changed from the version that shipped with test-10.\n\n### New Commands\n- **`apm add`** — add assistants to an existing installation without re-initializing.\n- **`apm remove`** — remove installed assistants. Destructive action confirmation, `--force` flag.\n- **`apm status`** — show current installation status (installed assistants, versions, source).\n- **`apm archive`** — session archival management. Create archives (snapshots all `.apm\u002F` contents), list with `--list`, delete with `--delete \u003Cname>`, clear all with `--clear`. Supports `--name \u003Cname>` for custom archive names.\n\n### Changed Behavior\n- **`apm init`** is now fresh-only — if APM is already initialized, it shows current status and points you to `apm add`, `apm update`, or `apm archive` instead of re-initializing.\n- **`apm update`** now auto-archives before updating, pre-validates assistants against the target release, of","2026-03-13T23:15:02",{"id":171,"version":172,"summary_zh":173,"released_at":174},323808,"v1.0.0-test-10","Outcome status simplification, Manager coordination efficiency, and behavioral fixes across guides and skills.\n\n## Changes\n\n- **Blocked status removed** — Outcome statuses reduced to Success, Partial, Failed. `failure_point` simplified to free-text `\u003Cdescription>` for all non-Success. Failed broadened to \"objective not achieved.\"\n- **Manager Tracker efficiency** — Review-dispatch cycle changes batched into a single Tracker edit. VC section static; branch state in task rows. Ready skipped when unblocked and dispatched in the same turn.\n- **Memory Notes deferred** — Durable observations accumulate as Working Notes, distilled into Memory Notes at Stage summary time. Stage summaries rewritten as prose + log list.\n- **Bus write safety** — Read-before-write on Task Prompt and Task Report delivery.\n- **Report scanning** — `\u002Fapm-5-check-reports` accepts multiple agent arguments; no-argument mode reads active dispatch buses with `ls` health check.\n- **Worker iteration guidance** — *Task Iteration* prompt body field: debug subagent after 3 failed attempts, report Partial if unresolved. Replaces framework rule in work-breakdown.\n- **Subagent verification** — Planner distinguishes detail-bearing findings (verify via handles) from general understanding (accept without verification).","2026-03-09T12:18:30",{"id":176,"version":177,"summary_zh":178,"released_at":179},323809,"v1.0.0-test-9","Behavioral fixes targeting the most persistent workflow issues and Copilot-specific regressions from `v1.0.0-test-8`.\n\n## Changes\n\n- **Per-stage write cycle enforcement** — Stage Cycles restructured as an explicit loop with write-first step and loop control. WORKFLOW.md §6.2 anchors per-stage writing at the spec level.\n- **Stage sequentiality** — New design principle: stages are sequential, parallel work uses intra-stage Task dispatch. Added to both WORKFLOW.md and work-breakdown §2.1.\n- **Framework rules** — APM_STANDARDS block template now includes two pre-populated rules: post-compaction procedural doc re-read, and debugging escalation over session exhaustion.\n- **Understanding summary guard** — Work structure signals in context-gathering now guard against premature decomposition language.\n- **Manager fixes** — Factual correction trigger in task-review content review, bulleted list format for Working\u002FMemory Notes, VC workspace identification reinforcement.\n- **Rules quality signal** — Classification dimension in Rules Analysis strengthened to signal that few genuinely universal rules exist per project.","2026-03-07T21:36:00",{"id":181,"version":182,"summary_zh":183,"released_at":184},323810,"v1.0.0-test-8","This testing preview is a **major update** from `v1.0.0-test-7`, with a comprehensive internal redesign of artifact naming, Memory architecture, file path conventions, and cross-file standards compliance. Every template file has been touched.\n\nCompared to the previous preview, this release focuses on making APM's internal system cleaner and more predictable: shorter artifact names, structured YAML schemas, kebab-case paths, and a clearer separation between live coordination state and durable project memory.\n\n## What improves for users (high impact)\n\n- **Simpler, shorter artifact names throughout**\n  - `Specifications` → `Spec`, `Implementation Plan` → `Plan`, `Execution Standards` → `Rules`. Better UX. Simpler and minimalist terminology creating a cleaner APM ecosystem.\n  \n- **Cleaner file system layout**\n  - All `.apm\u002F` paths are now kebab-case with zero-padded numbers. Files like `Implementation_Plan.md` and `Memory\u002FMemory_Root.md` become `plan.md` and `tracker.md`. Log files use `.log.md` suffix for clear identification: `.apm\u002Fmemory\u002Fstage-01\u002Ftask-01-02.log.md`.\n\n- **Memory architecture split: Tracker + Index**\n  - The old Memory Root has been split into two purpose-built documents. The **Tracker** (`tracker.md`) holds live project state — task tracking, agent tracking, VC state, and ephemeral working notes. The **Index** (`memory\u002Findex.md`) holds durable project memory — persistent observations, patterns, and stage summaries that survive Handoffs.\n\n- **Bus files simplified**\n  - `apm-task.md` → `task.md`, `apm-report.md` → `report.md`, `apm-handoff.md` → `handoff.md`. The `apm-` prefix was redundant inside the `.apm\u002Fbus\u002F` directory.\n\n- **Structured YAML frontmatter with field descriptions**\n  - Task Prompts, Task Logs, Handoff Logs, and batch envelopes now have documented YAML schemas with concrete examples and per-field descriptions. `agent_id` → `agent`, `memory_log_path` → `log_path`, `task_id` replaced by `stage`\u002F`task`\u002F`title` fields.\n\n## What improves for agents (behavioral)\n\n- **Communication protocol elevated to skill**\n  - Agent-to-user communication, visible reasoning, and terminology boundaries are now consolidated in the communication skill as the single source of truth. Guides no longer duplicate these instructions — they cross-reference instead.\n\n- **Formal note-taking system**\n  - New §2.6 in the Task Review guide distinguishes ephemeral Working Notes (coordination context in the Tracker, inserted and removed as needed) from durable Memory Notes (lasting patterns and insights in the Index).\n\n- **Manager Handoff prompts expanded for outstanding Tasks**\n  - The outgoing Manager now describes outstanding Tasks with enough detail for the incoming Manager to coordinate and review — objectives, instructions, review criteria, dependency context, workspace information. This is defensive: the clear-on-return protocol means Workers may have already cleared their Task Bus, so the original Task Prompt is gone by Handoff time. The incoming Manager needs the substance of what was dispatched, not just pointers.\n\n- **Handoff Logs restructured with proper YAML schemas**\n  - Both Manager and Worker Handoff Logs now use structured YAML frontmatter (`agent`, `outgoing_session`, `incoming_session`, `handoff`, `stage`) with documented field descriptions and reorganized body sections.\n\n- **Standards compliance across all files**\n  - 60+ compliance findings from a systematic audit have been addressed: reference format standardization, procedure transition naturalization, terminology consistency, formatting fixes, de-duplication, and cross-file behavioral alignment.\n\n## Main changes\n\n### Artifact templates (new file layout)\n\n- **Deleted:** `Implementation_Plan.md`, `Specifications.md`, `Memory\u002FMemory_Root.md`\n- **Added:** `plan.md`, `spec.md`, `tracker.md`, `memory\u002Findex.md` — all with YAML frontmatter\n\n### Standards\n\n- `WORKFLOW.md`: full rewrite of §5 Memory for Tracker\u002FIndex architecture, all term renames, bus file name updates, context rebuilding order updated, bus initialization moved from Manager to Planner\n- `TERMINOLOGY.md`: all term definitions and file locations updated\n- `STRUCTURE.md`: new §8.4 APM Artifact Files, §8.5 Task Identifiers (N.M prose, integer YAML, zero-padded paths)\n- `WRITING.md`: procedure transition example updated\n\n### Commands (all 7 updated)\n\n- All artifact references, bus file paths, and YAML field names updated\n- `apm-1`: bus initialization moved here as §4 Planning Phase Completion\n- `apm-2`: session detection changed from Handoff Bus to Tracker template state, Manager session 1 now populates both Tracker and Index\n- `apm-6`, `apm-7`: Handoff Log YAML schemas with field descriptions, restructured body sections, eligibility gates removed\n\n### Guides (all 6 updated)\n\n- `task-review.md`: new §2.6 Note-Taking Standards, restructured Tracker\u002FIndex format specs, stage summaries moved to Index\n- `task-logging.md`: Task Log YAML schema restructured (stage\u002Ftask\u002Ftitle fields), Next Steps ","2026-03-06T20:58:57",{"id":186,"version":187,"summary_zh":188,"released_at":189},323811,"v1.0.0-test-7","Seventh testing preview of APM v1.0.0. Introduces decision-point reasoning as a core workflow principle and applies formatting standardization across all 17 template files.\n\n## Highlights\n\n- **Decision-point reasoning** added as a core principle in WORKFLOW.md - agents now state brief reasoning at procedural decision points before acting on conclusions, making decision logic visible and auditable\n- **Visible Reasoning sections** added to four guides (task-assignment, task-execution, task-logging, task-review), each with role-specific assessment requirements\n- **Workspace context universalized** in Task Prompts - branch names now included for all dispatches, not just parallel\n\n## Changes\n\n### Workflow spec\n- New core principle: decision-point reasoning\n- Worker completion sequence now includes explicit VC commit step before logging\n\n### Commands\n- Added decision-point reasoning operating rule to Planner, Manager, and Worker\n- Normalized \"Memory Logs\" to \"Task Memory Logs\" in handoff commands\n- Fixed section cross-references after guide renumbering\n\n### Guides\n- New \"Visible Reasoning\" (§2.1) in task-assignment, task-execution, task-logging, task-review\n- New \"Outputs\" section in context-gathering defining the understanding summary\n- Context resolution assessment step added to context-gathering finalization\n- Natural language communication rule added to task-assignment and task-review\n- Explicit commit step added to task-execution completion\n- Fresh dispatch assessment rule in task-assignment\n- Decomposition validation checkpoint in work-breakdown Stage Cycles\n- Validation approach justification (inclusion\u002Fexclusion rationale) in work-breakdown\n\n### Skills\n- apm-communication: normalized to canonical \"Task Reports\" terminology\n- apm-version-control: deduplicated branch\u002Fworktree specs via cross-references\n\n### Formatting (all files)\n- Bold-to-italic on labeled list items per WRITING.md\n- Removed extraneous blank lines between intros and procedure lists\n- Added trailing periods to numbered list items","2026-03-02T11:01:44",{"id":191,"version":192,"summary_zh":193,"released_at":194},323812,"v1.0.0-test-6","Sixth testing preview of APM v1.0.0. Targeted writing compliance and behavioral fixes across 10 template files.\n\n## Changes\n\n- Removed anti-pattern guardrails that over-specified instead of stating generic rules\n- Added \"Signals, not structure\" guiding principle to context-gathering, elevating decomposition prohibition to top-level principle\n- Tightened verbosity in handoff commands and content guidelines\n- Worker validation: added external resource request rule and debug subagent escalation path\n- Task assignment: added Stage slug derivation rule and Task Prompt scoping - no coordination-level concepts\n- Task review: added within-authority action rule, merge-before-readiness check, and Stage Summary completion criteria\n- Version control: added multi-repo workspace detection and APM terminology exclusion from git artifacts\n- Planning: added Worker-scoped validation criteria and workspace structure capture in Specifications","2026-02-26T22:20:09",{"id":196,"version":197,"summary_zh":198,"released_at":199},323813,"v1.0.0-test-5","This testing preview is a **major update** from `v1.0.0-test-4`, with large structural changes across `templates\u002F`, `docs\u002F`, and `build\u002F`.\r\n\r\nCompared to the previous preview, this release focuses on making APM sessions easier to run in practice: clearer control commands, stricter workflow routing, explicit inter-agent communication, and more actionable planning guidance.\r\n\r\nFor your conveniance in testing, ive updated some of the documentation suite to cover the most important and practical aspects of the updated workflow (see below).\r\n\r\n## What improves for users (high impact)\r\n\r\n- **Clearer control flow during multi-agent orchestration**\r\n  - The command surface is now organized as a 7-command sequence.\r\n  - Task\u002Freport checks are explicit (`\u002Fapm-4-check-tasks`, `\u002Fapm-5-check-reports`) so communication checkpoints are visible and auditable, while the markdown code block copy-paste workflow gets deprecated (see below).\r\n\r\n- **A Message Bus mechanism is now introduced in this release window**\r\n  - `v1.0.0-test-5` introduces a file-based Message Bus model for cross-session communication.\r\n  - The communication model now routes task dispatch, report return, and handoff signaling through bus files instead of copy-pasting markdown code blocks (finally).\r\n\r\n- **Enhanced execution model replaces fragmented instruction layout**\r\n  - All procedures are now centered in dedicated APM guides (`context-gathering`, `work-breakdown`, `task-assignment`, `task-execution`, `task-review`, `task-logging`).\r\n  - The new design reduces instruction scattering and improves consistency between Planner\u002FManager\u002FWorker behavior.\r\n  - Parallel and batch task assignment & execution now supported!\r\n \r\n- **Planning quality and context handling are more explicit**\r\n  - The context gathering conversational approach has been refined so that the planner proactively research the codebase to inform it's decisions and avoid asking redundant questions.\r\n  - All planning document creation has been improved and now the planner produces much higher-quality output.\r\n\r\n## Main changes\r\n\r\n### Commands: refactored and renumbered\r\n\r\n- **Added**\r\n  - `templates\u002Fcommands\u002Fapm-4-check-tasks.md`\r\n  - `templates\u002Fcommands\u002Fapm-5-check-reports.md`\r\n  - `templates\u002Fcommands\u002Fapm-6-handoff-manager.md`\r\n  - `templates\u002Fcommands\u002Fapm-7-handoff-worker.md`\r\n\r\n- **Removed (legacy layout)**\r\n  - `templates\u002Fcommands\u002Fapm-4-initiate-delegate.md`\r\n  - `templates\u002Fcommands\u002Fapm-5-handoff-manager.md`\r\n  - `templates\u002Fcommands\u002Fapm-6-handoff-worker.md`\r\n\r\n- **Updated**\r\n  - `templates\u002Fcommands\u002Fapm-1-initiate-planner.md`\r\n  - `templates\u002Fcommands\u002Fapm-2-initiate-manager.md`\r\n  - `templates\u002Fcommands\u002Fapm-3-initiate-worker.md`\r\n\r\n### Guides and skills: major topology change\r\n\r\n- **Added guides**\r\n  - `templates\u002Fguides\u002Fcontext-gathering.md`\r\n  - `templates\u002Fguides\u002Fwork-breakdown.md`\r\n  - `templates\u002Fguides\u002Ftask-assignment.md`\r\n  - `templates\u002Fguides\u002Ftask-execution.md`\r\n  - `templates\u002Fguides\u002Ftask-review.md`\r\n  - `templates\u002Fguides\u002Ftask-logging.md`\r\n\r\n- **Added focused procedural skills**\r\n  - `templates\u002Fskills\u002Fapm-communication\u002FSKILL.md`\r\n    - includes `templates\u002Fskills\u002Fapm-communication\u002Fapm-bus-integration.md` for non-APM agent integration into an APM session.\r\n  - `templates\u002Fskills\u002Fapm-version-control\u002FSKILL.md`\r\n\r\n- **Removed legacy skills (consolidated into new model)**\r\n  - `templates\u002Fskills\u002Fartifact-maintenance.md`\r\n  - `templates\u002Fskills\u002Fcontext-gathering.md`\r\n  - `templates\u002Fskills\u002Fdelegate-debug.md`\r\n  - `templates\u002Fskills\u002Fdelegate-research.md`\r\n  - `templates\u002Fskills\u002Fmemory-logging.md`\r\n  - `templates\u002Fskills\u002Fmemory-maintenance.md`\r\n  - `templates\u002Fskills\u002Ftask-assignment.md`\r\n  - `templates\u002Fskills\u002Ftask-execution.md`\r\n  - `templates\u002Fskills\u002Fwork-breakdown.md`\r\n\r\n### Standards and specification layer\r\n\r\n- `templates\u002F_standards\u002FWORKFLOW.md` added as a canonical workflow spec which now dictates all commands\u002Fguides\u002Fskills content.\r\n- `templates\u002F_standards\u002FTERMINOLOGY.md`, `STRUCTURE.md`, and `WRITING.md` were substantially refactored and enhanced which propagated into the commands\u002Fguides\u002Fskills.\r\n\r\n### Build and packaging adjustments\r\n\r\n- Build pipeline was updated to support the guide\u002Fskill topology and new placeholders\u002Frouting.\r\n- Platform bundle generation remains available for:\r\n  - GitHub Copilot\r\n  - Claude Code\r\n  - Gemini CLI\r\n  - Cursor\r\n  - OpenCode\r\n\r\n## Documentation status in this preview\r\n\r\n- Core v1 docs were expanded\u002Frefined to explain the current workflow and orchestration model.\r\n- `docs\u002FREADME.md` now explicitly distinguishes:\r\n  - v1 preview docs currently aligned with this branch\r\n  - v0.5.x docs pending full rewrite\r\n\r\n---\r\n\r\nThanks for testing this preview. Feedback on command flow, Message Bus ergonomics, and guide usability is especially useful before stable `v1.0.0` gets out.\r\n","2026-02-22T20:40:11"]