[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-joelhooks--swarm-tools":3,"tool-joelhooks--swarm-tools":61},[4,18,26,36,44,53],{"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":10,"last_commit_at":24,"category_tags":25,"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":27,"name":28,"github_repo":29,"description_zh":30,"stars":31,"difficulty_score":32,"last_commit_at":33,"category_tags":34,"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",149489,2,"2026-04-10T11:32:46",[14,13,35],"语言模型",{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":32,"last_commit_at":42,"category_tags":43,"status":17},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。",108322,"2026-04-10T11:39:34",[14,15,13],{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":32,"last_commit_at":50,"category_tags":51,"status":17},6121,"gemini-cli","google-gemini\u002Fgemini-cli","gemini-cli 是一款由谷歌推出的开源 AI 命令行工具，它将强大的 Gemini 大模型能力直接集成到用户的终端环境中。对于习惯在命令行工作的开发者而言，它提供了一条从输入提示词到获取模型响应的最短路径，无需切换窗口即可享受智能辅助。\n\n这款工具主要解决了开发过程中频繁上下文切换的痛点，让用户能在熟悉的终端界面内直接完成代码理解、生成、调试以及自动化运维任务。无论是查询大型代码库、根据草图生成应用，还是执行复杂的 Git 操作，gemini-cli 都能通过自然语言指令高效处理。\n\n它特别适合广大软件工程师、DevOps 人员及技术研究人员使用。其核心亮点包括支持高达 100 万 token 的超长上下文窗口，具备出色的逻辑推理能力；内置 Google 搜索、文件操作及 Shell 命令执行等实用工具；更独特的是，它支持 MCP（模型上下文协议），允许用户灵活扩展自定义集成，连接如图像生成等外部能力。此外，个人谷歌账号即可享受免费的额度支持，且项目基于 Apache 2.0 协议完全开源，是提升终端工作效率的理想助手。",100752,"2026-04-10T01:20:03",[52,13,15,14],"插件",{"id":54,"name":55,"github_repo":56,"description_zh":57,"stars":58,"difficulty_score":32,"last_commit_at":59,"category_tags":60,"status":17},4721,"markitdown","microsoft\u002Fmarkitdown","MarkItDown 是一款由微软 AutoGen 团队打造的轻量级 Python 工具，专为将各类文件高效转换为 Markdown 格式而设计。它支持 PDF、Word、Excel、PPT、图片（含 OCR）、音频（含语音转录）、HTML 乃至 YouTube 链接等多种格式的解析，能够精准提取文档中的标题、列表、表格和链接等关键结构信息。\n\n在人工智能应用日益普及的今天，大语言模型（LLM）虽擅长处理文本，却难以直接读取复杂的二进制办公文档。MarkItDown 恰好解决了这一痛点，它将非结构化或半结构化的文件转化为模型“原生理解”且 Token 效率极高的 Markdown 格式，成为连接本地文件与 AI 分析 pipeline 的理想桥梁。此外，它还提供了 MCP（模型上下文协议）服务器，可无缝集成到 Claude Desktop 等 LLM 应用中。\n\n这款工具特别适合开发者、数据科学家及 AI 研究人员使用，尤其是那些需要构建文档检索增强生成（RAG）系统、进行批量文本分析或希望让 AI 助手直接“阅读”本地文件的用户。虽然生成的内容也具备一定可读性，但其核心优势在于为机器",93400,"2026-04-06T19:52:38",[52,14],{"id":62,"github_repo":63,"name":64,"description_en":65,"description_zh":66,"ai_summary_zh":66,"readme_en":67,"readme_zh":68,"quickstart_zh":69,"use_case_zh":70,"hero_image_url":71,"owner_login":72,"owner_name":73,"owner_avatar_url":74,"owner_bio":75,"owner_company":76,"owner_location":77,"owner_email":78,"owner_twitter":72,"owner_website":79,"owner_url":80,"languages":81,"stars":106,"forks":107,"last_commit_at":108,"license":109,"difficulty_score":32,"env_os":110,"env_gpu":111,"env_ram":110,"env_deps":112,"category_tags":120,"github_topics":121,"view_count":32,"oss_zip_url":109,"oss_zip_packed_at":109,"status":17,"created_at":127,"updated_at":128,"faqs":129,"releases":159},6338,"joelhooks\u002Fswarm-tools","swarm-tools","🐝 Multi-agent swarm coordination for OpenCode with learning capabilities, agent issue tracking, and management","swarm-tools 是一款专为 OpenCode 和 Claude Code 设计的多智能体协作插件，旨在让 AI 像蜂群一样高效处理复杂编程任务。它核心解决了大模型在处理大型需求时容易“遗忘上下文”或陷入混乱的痛点，通过将宏大目标拆解为可并行执行的子任务，协调多个代理同时工作，并在会话中断后依然能保持进度不丢失。\n\n这款工具特别适合需要处理复杂功能开发、系统重构或大规模代码库维护的软件开发者。其独特亮点在于引入了三大核心机制：基于 Git 的\"Hive\"任务追踪系统，确保任务状态持久化且可同步；具备语义记忆能力的\"Hivemind\"，能让系统从过往成功实践中学习并优化策略；以及基于演员模型的\"Swarm Mail\"通信协议，通过文件预留机制完美避免多代理修改同一文件时的冲突。\n\n使用非常简单，只需输入 `\u002Fswarm` 加上任务描述（如“添加 OAuth 认证”），协调器便会自动规划策略、分配工人代理执行，并在完成后进行审核。swarm-tools 不仅提升了编码效率，更让多智能体协作变得可控、可靠且具备持续进化的能力，是构建高质量软件工程的得力助手。","# Swarm\n\n[![npm version](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fv\u002Fopencode-swarm-plugin.svg)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002Fopencode-swarm-plugin)\n[![Documentation](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-swarmtools.ai-blue)](https:\u002F\u002Fswarmtools.ai\u002Fdocs)\n\n```\n ███████╗██╗    ██╗ █████╗ ██████╗ ███╗   ███╗\n ██╔════╝██║    ██║██╔══██╗██╔══██╗████╗ ████║\n ███████╗██║ █╗ ██║███████║██████╔╝██╔████╔██║\n ╚════██║██║███╗██║██╔══██║██╔══██╗██║╚██╔╝██║\n ███████║╚███╔███╔╝██║  ██║██║  ██║██║ ╚═╝ ██║\n ╚══════╝ ╚══╝╚══╝ ╚═╝  ╚═╝╚═╝  ╚═╝╚═╝     ╚═╝\n```\n\n**Multi-agent coordination that survives context death.**\n\nBreak big tasks into small ones. Spawn agents to work in parallel. Learn from what works.\n\n> **[swarmtools.ai](https:\u002F\u002Fswarmtools.ai)** · **[Documentation](https:\u002F\u002Fswarmtools.ai\u002Fdocs)**\n\n---\n\n## Installation\n\n### OpenCode\n\n```bash\nnpm install -g opencode-swarm-plugin\nswarm setup\n```\n\nDone. Use `\u002Fswarm \"your task\"` in any OpenCode session.\n\n### Claude Code\n\n**Step 1:** Install the CLI globally (required):\n\n```bash\nnpm install -g opencode-swarm-plugin\n```\n\n**Step 2:** Add the marketplace in Claude Code:\n\n```\n\u002Fplugin\n→ Manage marketplaces\n→ Add marketplace\n→ Enter: joelhooks\u002Fswarm-tools\n```\n\n**Step 3:** Install the plugin:\n\n```\n\u002Fplugin\n→ Manage plugins\n→ swarm-tools\n→ swarm\n→ Install\n```\n\nThe MCP server starts automatically. Use `\u002Fswarm \"your task\"` in any session.\n\n---\n\n## What It Does\n\n```\n\u002Fswarm \"Add user authentication with OAuth\"\n```\n\n1. **Decomposes** the task into parallelizable subtasks\n2. **Creates cells** in the Hive (git-backed task tracker)\n3. **Spawns workers** with file reservations (no conflicts)\n4. **Coordinates** via Swarm Mail (embedded event store)\n5. **Reviews** each completion before approval\n6. **Learns** what worked for next time\n\n---\n\n## Core Concepts\n\n### Hive\n\nGit-backed task tracking in `.hive\u002F`. Survives sessions, syncs via git.\n\n```\nhive_create({ title: \"Fix auth bug\", type: \"bug\" })\nhive_cells({ status: \"in_progress\" })\nhive_close({ id: \"cell-123\", reason: \"Fixed\" })\n```\n\n### Hivemind\n\nSemantic memory with embeddings. Store learnings, search later.\n\n```\nhivemind_store({ information: \"Auth requires idempotency keys\", tags: \"auth,gotcha\" })\nhivemind_find({ query: \"auth patterns\" })\n```\n\n### Swarm Mail\n\nActor-model coordination between agents. File reservations, messaging, checkpoints.\n\n```\nswarmmail_reserve({ paths: [\"src\u002Fauth\u002F*\"], exclusive: true })\nswarmmail_send({ to: [\"worker-b\"], subject: \"Need types\", body: \"...\" })\n```\n\n---\n\n## Commands\n\n| Command | Description |\n|---------|-------------|\n| `\u002Fswarm \u003Ctask>` | Decompose and spawn parallel workers |\n| `\u002Fhive` | Query and manage tasks |\n| `\u002Finbox` | Check messages from other agents |\n| `\u002Fstatus` | Swarm coordination status |\n| `\u002Fhandoff` | End session with sync and handoff notes |\n\n---\n\n## CLI\n\n```bash\nswarm setup      # Configure OpenCode\u002FClaude Code integration\nswarm doctor     # Check dependencies (Ollama for embeddings)\nswarm init       # Initialize hive in current project\nswarm config     # Show config paths\n```\n\n---\n\n## How It Works\n\n```\n                         Your Task\n                             │\n                             ▼\n                    ┌────────────────┐\n                    │  COORDINATOR   │\n                    │                │\n                    │ 1. Query past  │\n                    │    sessions    │\n                    │ 2. Pick strat  │\n                    │ 3. Decompose   │\n                    └────────────────┘\n                             │\n           ┌─────────────────┼─────────────────┐\n           ▼                 ▼                 ▼\n    ┌────────────┐    ┌────────────┐    ┌────────────┐\n    │  Worker A  │    │  Worker B  │    │  Worker C  │\n    │            │    │            │    │            │\n    │ 🔒 files   │───▶│ 🔒 files   │    │ 🔒 files   │\n    └────────────┘    └────────────┘    └────────────┘\n           │                 │                 │\n           └─────────────────┼─────────────────┘\n                             ▼\n                    ┌────────────────┐\n                    │   LEARNING     │\n                    │                │\n                    │ Record outcome │\n                    │ Update weights │\n                    └────────────────┘\n```\n\n---\n\n## Architecture\n\nEverything runs locally. No external servers.\n\n- **libSQL** - Embedded SQLite for event sourcing\n- **Hive** - Git-backed `.hive\u002F` directory for tasks\n- **Hivemind** - Semantic memory with Ollama embeddings (falls back to FTS)\n- **Swarm Mail** - DurableMailbox, DurableLock, DurableDeferred primitives\n\n### Event Sourcing\n\nAll state is an append-only event log:\n\n```\nagent_registered → Agent joins swarm\nmessage_sent     → Agent-to-agent communication\nfile_reserved    → Exclusive lock acquired\nfile_released    → Lock released\ncheckpoint       → Progress snapshot\noutcome          → Completion result\n```\n\n---\n\n## Learning System\n\nEvery completion records duration, errors, files touched, success.\n\n- **Patterns** mature: candidate → established → proven\n- **Anti-patterns** auto-generate when failure rate > 60%\n- **Confidence decays** over 90 days unless revalidated\n\n---\n\n## Dependencies\n\n| Required | Optional |\n|----------|----------|\n| [Bun](https:\u002F\u002Fbun.sh) | [Ollama](https:\u002F\u002Follama.ai) - local embeddings |\n\nRun `swarm doctor` to check status.\n\n### Embedding Model Configuration\n\nConfigure the embedding model via environment variables:\n\n```bash\nexport OLLAMA_MODEL=nomic-embed-text  # Default: mxbai-embed-large\nexport OLLAMA_HOST=http:\u002F\u002Flocalhost:11434  # Default\n```\n\nSupported models: `mxbai-embed-large` (1024d), `nomic-embed-text` (768d), `all-minilm` (384d), `snowflake-arctic-embed` (1024d). See [swarm-mail docs](packages\u002Fswarm-mail\u002FREADME.md#configuring-the-embedding-model) for details.\n\n---\n\n## Development\n\n```bash\nbun install\nbun turbo build\nbun turbo test\n```\n\n---\n\n## Credits\n\n- [MCP Agent Mail](https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fmcp_agent_mail) - inspiration for multi-agent coordination\n- [Electric SQL](https:\u002F\u002Felectric-sql.com) - durable streams patterns\n- [Superpowers](https:\u002F\u002Fgithub.com\u002Fobra\u002Fsuperpowers) - verification patterns\n\n---\n\nMIT\n","# 群体\n\n[![npm version](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fv\u002Fopencode-swarm-plugin.svg)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002Fopencode-swarm-plugin)\n[![Documentation](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdocs-swarmtools.ai-blue)](https:\u002F\u002Fswarmtools.ai\u002Fdocs)\n\n```\n ███████╗██╗    ██╗ █████╗ ██████╗ ███╗   ███╗\n ██╔════╝██║    ██║██╔══██╗██╔══██╗████╗ ████║\n ███████╗██║ █╗ ██║███████║██████╔╝██╔████╔██║\n ╚════██║██║███╗██║██╔══██║██╔══██╗██║╚██╔╝██║\n ███████║╚███╔███╔╝██║  ██║██║  ██║██║ ╚═╝ ██║\n ╚══════╝ ╚══╝╚══╝ ╚═╝  ╚═╝╚═╝  ╚═╝╚═╝     ╚═╝\n```\n\n**多智能体协调，即使上下文失效也能持续运行。**\n\n将大任务拆解为小任务。并行启动多个智能体协同工作。从成功经验中学习。\n\n> **[swarmtools.ai](https:\u002F\u002Fswarmtools.ai)** · **[文档](https:\u002F\u002Fswarmtools.ai\u002Fdocs)**\n\n---\n\n## 安装\n\n### OpenCode\n\n```bash\nnpm install -g opencode-swarm-plugin\nswarm setup\n```\n\n完成。在任何 OpenCode 会话中使用 `\u002Fswarm \"你的任务\"`。\n\n### Claude Code\n\n**步骤 1：** 全局安装 CLI（必需）：\n\n```bash\nnpm install -g opencode-swarm-plugin\n```\n\n**步骤 2：** 在 Claude Code 中添加市场：\n\n```\n\u002Fplugin\n→ 管理市场\n→ 添加市场\n→ 输入：joelhooks\u002Fswarm-tools\n```\n\n**步骤 3：** 安装插件：\n\n```\n\u002Fplugin\n→ 管理插件\n→ swarm-tools\n→ swarm\n→ 安装\n```\n\nMCP 服务器会自动启动。在任何会话中使用 `\u002Fswarm \"你的任务\"`。\n\n---\n\n## 功能简介\n\n```\n\u002Fswarm \"使用 OAuth 添加用户认证\"\n```\n\n1. **分解**任务为可并行执行的子任务\n2. **创建蜂巢单元**（基于 Git 的任务跟踪器）\n3. **启动工作者**并预留文件（避免冲突）\n4. **通过 Swarm Mail 协调**（嵌入式事件存储）\n5. **审查**每次完成情况后再批准\n6. **学习**哪些方法有效，以便下次改进\n\n---\n\n## 核心概念\n\n### 蜂巢\n\n基于 Git 的任务跟踪系统，位于 `.hive\u002F` 目录下。跨会话持久化，并通过 Git 同步。\n\n```\nhive_create({ title: \"修复认证漏洞\", type: \"bug\" })\nhive_cells({ status: \"in_progress\" })\nhive_close({ id: \"cell-123\", reason: \"已修复\" })\n```\n\n### 蜂群心智\n\n带有嵌入向量的语义记忆。存储学习内容，便于后续检索。\n\n```\nhivemind_store({ information: \"认证需要幂等密钥\", tags: \"auth,gotcha\" })\nhivemind_find({ query: \"认证模式\" })\n```\n\n### Swarm Mail\n\n基于 Actor 模型的智能体间协调机制。包括文件预留、消息传递和检查点。\n\n```\nswarmmail_reserve({ paths: [\"src\u002Fauth\u002F*\"], exclusive: true })\nswarmmail_send({ to: [\"worker-b\"], subject: \"需要类型\", body: \"...\" })\n```\n\n---\n\n## 命令\n\n| 命令 | 描述 |\n|------|------|\n| `\u002Fswarm \u003Ctask>` | 分解任务并启动并行工作者 |\n| `\u002Fhive` | 查询和管理任务 |\n| `\u002Finbox` | 查看其他智能体的消息 |\n| `\u002Fstatus` | 显示 Swarm 协调状态 |\n| `\u002Fhandoff` | 结束会话并同步手记 |\n\n---\n\n## CLI\n\n```bash\nswarm setup      # 配置 OpenCode\u002FClaude Code 集成\nswarm doctor     # 检查依赖项（Ollama 用于嵌入）\nswarm init       # 初始化当前项目的蜂巢\nswarm config     # 显示配置路径\n```\n\n---\n\n## 工作原理\n\n```\n                         你的任务\n                             │\n                             ▼\n                    ┌────────────────┐\n                    │  协调者      │\n                    │                │\n                    │ 1. 查询历史  │\n                    │    会话      │\n                    │ 2. 选择策略  │\n                    │ 3. 分解      │\n                    └────────────────┘\n                             │\n           ┌─────────────────┼─────────────────┐\n           ▼                 ▼                 ▼\n    ┌────────────┐    ┌────────────┐    ┌────────────┐\n    │  Worker A  │    │  Worker B  │    │  Worker C  │\n    │            │    │            │    │            │\n    │ 🔒 文件   │───▶│ 🔒 文件   │    │ 🔒 文件   │\n    └────────────┘    └────────────┘    └────────────┘\n           │                 │                 │\n           └─────────────────┼─────────────────┘\n                             ▼\n                    ┌────────────────┐\n                    │   学习中心    │\n                    │                │\n                    │ 记录结果     │\n                    │ 更新权重     │\n                    └────────────────┘\n```\n\n---\n\n## 架构\n\n所有组件均在本地运行，无需外部服务器。\n\n- **libSQL** - 嵌入式 SQLite 用于事件溯源\n- **蜂巢** - 基于 Git 的 `.hive\u002F` 目录用于任务管理\n- **蜂群心智** - 使用 Ollama 嵌入的语义记忆（回退至 FTS）\n- **Swarm Mail** - 提供 DurableMailbox、DurableLock 和 DurableDeferred 原语\n\n### 事件溯源\n\n所有状态以追加方式记录在事件日志中：\n\n```\nagent_registered → 智能体加入 Swarm\nmessage_sent     → 智能体间通信\nfile_reserved    → 获取独占锁\nfile_released    → 释放锁\ncheckpoint       → 进度快照\noutcome          → 完成结果\n```\n\n---\n\n## 学习系统\n\n每次完成都会记录耗时、错误、涉及的文件以及是否成功。\n\n- **模式**会逐步成熟：候选 → 稳定 → 成熟\n- **反模式**会在失败率超过 60% 时自动生成\n- **信心值**会在 90 天内逐渐衰减，除非重新验证\n\n---\n\n## 依赖项\n\n| 必需 | 可选 |\n|------|------|\n| [Bun](https:\u002F\u002Fbun.sh) | [Ollama](https:\u002F\u002Follama.ai) - 本地嵌入 |\n\n运行 `swarm doctor` 检查状态。\n\n### 嵌入模型配置\n\n通过环境变量配置嵌入模型：\n\n```bash\nexport OLLAMA_MODEL=nomic-embed-text  # 默认：mxbai-embed-large\nexport OLLAMA_HOST=http:\u002F\u002Flocalhost:11434  # 默认\n```\n\n支持的模型：`mxbai-embed-large` (1024d), `nomic-embed-text` (768d), `all-minilm` (384d), `snowflake-arctic-embed` (1024d)。详情请参阅 [swarm-mail 文档](packages\u002Fswarm-mail\u002FREADME.md#configuring-the-embedding-model)。\n\n---\n\n## 开发\n\n```bash\nbun install\nbun turbo build\nbun turbo test\n```\n\n---\n\n## 致谢\n\n- [MCP Agent Mail](https:\u002F\u002Fgithub.com\u002FDicklesworthstone\u002Fmcp_agent_mail) - 多智能体协调的灵感来源\n- [Electric SQL](https:\u002F\u002Felectric-sql.com) - 耐久流模式\n- [Superpowers](https:\u002F\u002Fgithub.com\u002Fobra\u002Fsuperpowers) - 验证模式\n\n---\n\nMIT","# Swarm Tools 快速上手指南\n\nSwarm 是一个专为多智能体协作设计的开源工具，旨在解决上下文丢失问题。它能将大型任务拆解为可并行执行的子任务，自动调度多个 Agent 协同工作，并通过本地持久化存储（Hive）和语义记忆（Hivemind）从历史执行中学习优化。\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*   **运行时依赖**：\n    *   **Bun** (必需)：高性能 JavaScript 运行时。\n        ```bash\n        curl -fsSL https:\u002F\u002Fbun.sh\u002Finstall | bash\n        ```\n        *(国内用户若下载缓慢，可尝试使用镜像源或手动下载安装包)*\n    *   **Node.js\u002Fnpm** (可选但推荐)：用于全局安装插件 CLI。\n*   **AI 模型依赖** (可选但强烈推荐)：\n    *   **Ollama**：用于本地运行嵌入模型（实现语义记忆和任务检索）。\n        ```bash\n        curl -fsSL https:\u002F\u002Follama.com\u002Finstall.sh | sh\n        ollama pull mxbai-embed-large\n        ```\n        *(注：默认模型为 `mxbai-embed-large`，也可配置为 `nomic-embed-text` 等)*\n\n## 安装步骤\n\n根据您使用的代码助手平台，选择以下一种方式进行安装：\n\n### 方案 A：OpenCode 用户\n\n1.  **全局安装插件**：\n    ```bash\n    npm install -g opencode-swarm-plugin\n    ```\n2.  **初始化配置**：\n    ```bash\n    swarm setup\n    ```\n3.  **完成**：安装成功后即可在任何 OpenCode 会话中使用。\n\n### 方案 B：Claude Code 用户\n\n1.  **全局安装 CLI** (必需步骤)：\n    ```bash\n    npm install -g opencode-swarm-plugin\n    ```\n\n2.  **添加市场源**：\n    在 Claude Code 中输入 `\u002Fplugin`，依次选择：\n    *   `Manage marketplaces`\n    *   `Add marketplace`\n    *   输入地址：`joelhooks\u002Fswarm-tools`\n\n3.  **安装插件**：\n    再次输入 `\u002Fplugin`，依次选择：\n    *   `Manage plugins`\n    *   找到 `swarm-tools` -> `swarm`\n    *   点击 `Install`\n\n4.  **完成**：MCP 服务器将自动启动，可在任意会话中调用。\n\n## 基本使用\n\n安装完成后，无需复杂配置，直接在对话中通过 `\u002Fswarm` 命令即可触发多智能体协作流程。\n\n### 1. 执行并行任务\n输入以下命令，Swarm 会自动拆解任务、分配文件锁并调度多个 Worker 并行处理：\n\n```text\n\u002Fswarm \"Add user authentication with OAuth\"\n```\n\n**执行流程预览：**\n*   **拆解**：将任务分解为可并行的子任务。\n*   **追踪**：在 `.hive\u002F` 目录创建 Git 支持的任务单元格。\n*   **锁定**：为每个 Worker 预留文件路径，避免冲突。\n*   **协作**：通过 Swarm Mail 进行代理间通信。\n*   **审核**：在完成前审查每个子任务结果。\n*   **学习**：记录成功模式以供未来参考。\n\n### 2. 常用管理命令\n\n| 命令 | 功能描述 |\n| :--- | :--- |\n| `\u002Fswarm \u003Ctask>` | 拆解任务并生成并行 Worker |\n| `\u002Fhive` | 查询和管理任务状态 |\n| `\u002Finbox` | 查看来自其他 Agent 的消息 |\n| `\u002Fstatus` | 查看当前协调状态 |\n| `\u002Fhandoff` | 结束会话并同步交接笔记 |\n\n### 3. 本地 CLI 工具\n除了聊天命令外，您还可以在终端直接使用以下工具进行维护：\n\n```bash\nswarm doctor     # 检查依赖状态（如 Ollama 是否就绪）\nswarm init       # 在当前项目初始化 Hive 跟踪系统\nswarm config     # 显示配置文件路径\n```\n\n> **提示**：所有数据（任务记录、语义记忆、事件日志）均存储在本地 `.hive\u002F` 目录和嵌入式 SQLite 数据库中，无需连接外部服务器，确保数据安全与隐私。","某初创团队需要在两天内为现有的电商后端紧急集成 OAuth 登录功能，同时保证现有代码库的稳定性。\n\n### 没有 swarm-tools 时\n- **上下文断裂风险高**：开发者在分拆任务给不同成员或多次会话时，常因会话重启导致“上下文死亡”，需反复重新解释架构和进度。\n- **文件冲突频发**：多人或并行修改 `src\u002Fauth\u002F` 目录下的核心文件时，极易产生 Git 合并冲突，排查耗时且容易引入回归错误。\n- **经验无法沉淀**：解决过的鉴权坑点（如幂等性密钥处理）仅存在于个人笔记或聊天记录中，下次遇到类似问题无法被系统自动检索复用。\n- **协作状态黑盒**：缺乏统一的实时看板，管理者难以知晓哪个子任务卡住、谁在等待资源，协调沟通成本极高。\n\n### 使用 swarm-tools 后\n- **会话持久化追踪**：swarm-tools 利用基于 Git 的 Hive 机制记录所有任务单元格，即使会话中断或重启，进度和上下文也能完美同步恢复。\n- **智能文件锁机制**：通过 Swarm Mail 自动预留文件路径，Worker A 和 Worker B 在并行开发时互不干扰，从根源上杜绝了代码冲突。\n- **群体智慧自学习**：Hivemind 组件自动将本次修复的\"Auth 幂等性”方案向量化存储，未来遇到相似需求时可立即检索并推荐最佳实践。\n- **自动化协同流转**：Coordinator 自动拆解任务并分发，各 Worker 完成后自动触发审查流程，全流程状态通过 `\u002Fstatus` 一目了然。\n\nswarm-tools 通过将复杂的多人协作转化为可持久化、自学习的多智能体并行流，让大型功能开发像单兵作战一样流畅高效。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fjoelhooks_swarm-tools_da8619e2.png","joelhooks","Joel Hooks","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fjoelhooks_3a4b285e.jpg","working on @skillrecordings building @badass-courses like @ai-hero-dev and @total-typescript ","egghead.io","Vancouver, WA","joelhooks@gmail.com","http:\u002F\u002Fbadass.dev","https:\u002F\u002Fgithub.com\u002Fjoelhooks",[82,86,90,94,98,102],{"name":83,"color":84,"percentage":85},"TypeScript","#3178c6",86,{"name":87,"color":88,"percentage":89},"JavaScript","#f1e05a",9.5,{"name":91,"color":92,"percentage":93},"MDX","#fcb32c",4.2,{"name":95,"color":96,"percentage":97},"Shell","#89e051",0.2,{"name":99,"color":100,"percentage":101},"CSS","#663399",0.1,{"name":103,"color":104,"percentage":105},"HTML","#e34c26",0,621,51,"2026-04-10T13:14:06",null,"未说明","非必需。若需本地语义记忆功能，可选配支持 Ollama 的硬件（通常依赖 CPU 或本地 GPU 运行嵌入模型），无具体型号要求。",{"notes":113,"python":114,"dependencies":115},"该工具基于 Node.js 生态，使用 Bun 作为主要运行时，而非 Python。核心功能完全在本地运行，无需外部服务器。若启用‘蜂群思维’（Hivemind）语义记忆功能，需安装并运行 Ollama 服务，默认使用 mxbai-embed-large 模型，也可配置为 nomic-embed-text 等其他模型。文件锁和任务状态通过 Git 和嵌入式 SQLite (libSQL) 管理。","不需要 Python",[116,117,118,119],"Bun (运行时必需)","Ollama (可选，用于本地嵌入模型)","libSQL (内置)","Git (用于 Hive 任务追踪)",[13,14],[122,123,124,125,126],"ai-agents","machine-learning","multi-agent","opencode","swarm","2026-03-27T02:49:30.150509","2026-04-11T03:24:41.196290",[130,135,140,145,150,155],{"id":131,"question_zh":132,"answer_zh":133,"source_url":134},28697,"为什么会出现 'semantic-memory_find' 工具不可用的错误？","这通常由两个原因导致：1. Ollama 连接失败时，系统未能优雅地降级到全文搜索（FTS）；2. 维度不匹配错误（如 0 != 1024），这是因为 Ollama 不可用时存储的记忆没有嵌入向量。解决方案是升级到 v0.43.0 或更高版本，该版本修复了 FTS 回退机制，并引入了 `semantic-memory_repair` 工具来清理和重新嵌入过时的记忆数据。","https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fissues\u002F53",{"id":136,"question_zh":137,"answer_zh":138,"source_url":139},28698,"运行 'swarm init' 时在 'Create your first cell?' 提示处卡住无法输入怎么办？","这可能是终端状态重置问题（尤其在 Windows 上）。解决方法包括：1. 尝试使用 `--verbose` 标志运行以获取更多细节；2. 安装最新的 beta 版本（如 `bun add -g @primeinc\u002Fswarm@0.57.9-beta.0` 或 `npm install -g @primeinc\u002Fswarm@beta`），其中包含了针对 Windows 和 SQL 的修复；3. 如果是源码运行，可在 `swarm.ts` 文件的 spinner 停止后、确认提示前添加 `await new Promise(r => setImmediate(r));` 以让出事件循环重置终端状态。","https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fissues\u002F138",{"id":141,"question_zh":142,"answer_zh":143,"source_url":144},28699,"执行 swarm 命令时报错 'Cannot find module evalite\u002Frunner' 如何解决？","这是一个依赖缺失问题。临时解决方法是全局安装缺失的依赖包：运行命令安装 `evalite` 以及其级联依赖 `vitest`（例如 `npm install -g evalite vitest`）。建议同时运行 `swarm doctor` 诊断问题，并确保升级到最新版本以获取官方修复。","https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fissues\u002F95",{"id":146,"question_zh":147,"answer_zh":148,"source_url":149},28700,"Swarm 是否支持 Git Worktrees？多个 worktree 会共享数据库吗？","在 v0.43.0 之前，每个 worktree 会写入独立的数据库，导致状态不同步。从 v0.43.0 开始已修复此问题：所有 worktree 现在共享主仓库的数据库。系统会自动检测是否在 worktree 中（通过检查 `.git` 是文件还是目录并读取 `gitdir` 路径），并解析到主仓库的 DB 位置，确保记忆、邮件等数据在所有 worktree 间同步。","https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fissues\u002F52",{"id":151,"question_zh":152,"answer_zh":153,"source_url":154},28701,"Markdown 输出中出现奇怪符号或乱码怎么办？","这通常是本地区域设置（Locale）或字符映射配置问题。请检查终端环境：1. 运行 `echo $LANG` 确认是否为 UTF-8（如 en_US.UTF-8）；2. 运行 `locale charmap` 检查字符映射；3. 运行 `locale` 查看所有设置，确保 `LC_CTYPE` 等相关变量均为 UTF-8。如果在 WSL2 或 Ubuntu 中遇到此问题，即使显示为 UTF-8，也可能需要重启终端或重新配置 locale 生成。","https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fissues\u002F129",{"id":156,"question_zh":157,"answer_zh":158,"source_url":134},28702,"当 Ollama 服务不可用时，语义记忆搜索会如何处理？","在旧版本中，Ollama 不可用会导致直接报错。在新版本（v0.43.0+）中，系统实现了优雅的降级机制：当检测到 Ollama 连接失败时，会自动静默回退到全文搜索（FTS），不再抛出错误。如果之前因 Ollama 宕机存入了无嵌入向量的脏数据，可以使用新增的 `semantic-memory_repair` 工具进行清理和重新嵌入。",[160,165,170,175,180,185,190,194,199,204,209,214,219,224,229,233,238,243,248,253],{"id":161,"version":162,"summary_zh":163,"released_at":164},197594,"claude-code-swarm-plugin@0.63.3","### 补丁变更\n\n-   [`f21f581`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002Ff21f581c2efeb1407ae2761bfb2f4a641714e212) 感谢 [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - 杂项：链接插件版本 — 将 claude-code-swarm-plugin 与 opencode-swarm-plugin 同步\n","2026-02-06T17:18:31",{"id":166,"version":167,"summary_zh":168,"released_at":169},197595,"@swarmtools\u002Fevals@0.2.40","### 补丁变更\n\n-   更新了依赖项 \\[[`35a81f3`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F35a81f38318ed08b835e7269405e8e53ed4599a5)]:\n    -   opencode-swarm-plugin@0.63.2\n","2026-02-06T17:04:49",{"id":171,"version":172,"summary_zh":173,"released_at":174},197596,"opencode-swarm-plugin@0.63.2","### 补丁变更\n\n-   [`35a81f3`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F35a81f38318ed08b835e7269405e8e53ed4599a5) 感谢 [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - 修复（发布）：将 Bun 升级至 1.3.8，并添加工作区依赖解析的安全保障。\n\n    此前的修复版本（0.63.1）仍然会打包未解析的 `workspace:*` 依赖，因为 CI 通过 `packageManager` 固定了 Bun 的版本为 1.3.4。本次更新将 Bun 升级至 1.3.8，并用一个更完善的发布脚本替换了原有的内联单行命令。该脚本会在上传到 npm 之前，验证并解析所有可能遗漏的 `workspace:*` 引用。","2026-02-06T17:04:47",{"id":176,"version":177,"summary_zh":178,"released_at":179},197597,"@swarmtools\u002Fevals@0.2.39","### 补丁变更\n\n-   更新了依赖项 \\[[`d87cc94`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002Fd87cc9467edfcb38c832b752b81fc7118d47dd38)]:\n    -   opencode-swarm-plugin@0.63.1\n","2026-02-06T16:57:21",{"id":181,"version":182,"summary_zh":183,"released_at":184},197598,"opencode-swarm-plugin@0.63.1","### 补丁变更\n\n-   [`d87cc94`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002Fd87cc9467edfcb38c832b752b81fc7118d47dd38) 感谢 [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - 修复 (发布): 在执行 npm 发布前解析 workspace:\\* 依赖项\n\n    `bun publish v1.3.4` 会静默地将未解析的 `workspace:*` 依赖项发布到 npm，\n    导致 `opencode-swarm-plugin@0.63.0` 的安装失败。现将 CI 发布流程切换为\n    `bun pm pack`（该命令会正确解析 workspace 协议）加上 `npm publish \u003Ctarball>`。\n\n    > “我认为任何未解决的 bug 都是对个人尊严的不可容忍的侮辱”\n    > — 约翰·欧斯特豪特，《软件设计哲学》","2026-02-06T16:57:18",{"id":186,"version":187,"summary_zh":188,"released_at":189},197599,"claude-code-swarm-plugin@0.61.0","### 小幅改动\n\n-   feat: 改进 swarm 工具的模式和文档\n\n    -   修复 MCP 数组参数模式（files、paths、to、files_touched 等）——使用正确的 JSON Schema 数组类型，而非仅以“JSON 数组”描述的字符串\n    -   在 swarm 协作技能中添加明确的生成示例，展示正确的 swarm_spawn_subtask 格式\n    -   为 ralph 任务模式添加 ready_for_review 状态，以支持规范的评审流程\n    -   修正技能的 frontmatter，使其仅使用 name 和 description 字段，符合技能创建者规范\n    -   从插件分发中移除 release 技能（仅适用于项目特定场景）","2026-02-06T16:43:25",{"id":191,"version":192,"summary_zh":188,"released_at":193},197600,"opencode-swarm-plugin@0.63.0","2026-02-06T16:43:22",{"id":195,"version":196,"summary_zh":197,"released_at":198},197601,"@swarmtools\u002Fevals@0.2.38","### 补丁变更\n\n-   更新了依赖项 \\[]：\n    -   opencode-swarm-plugin@0.63.0\n","2026-02-06T16:43:19",{"id":200,"version":201,"summary_zh":202,"released_at":203},197602,"@swarmtools\u002Fevals@0.2.37","### 补丁变更\n\n-   更新了依赖项 \\[[`765e442`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F765e442407ac9d9905481460bc57192db69e4283), [`453fe90`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F453fe9063070a483dc64bfaabbf99362c8674e48)]:\n    -   swarm-mail@1.11.2\n    -   opencode-swarm-plugin@0.62.2\n","2026-02-06T02:00:15",{"id":205,"version":206,"summary_zh":207,"released_at":208},197603,"claude-code-swarm-plugin@0.60.2","### 补丁变更\n\n-   [`453fe90`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F453fe9063070a483dc64bfaabbf99362c8674e48) 感谢 [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - 修复（版本）：通过 changesets 生命周期钩子同步所有 plugin.json 清单文件\n\n    之前 plugin.json 文件从未被 changesets 更新过，导致版本出现偏差：\n\n    -   opencode-swarm-plugin 的 plugin.json 停留在 0.59.5（package.json 为 0.62.0）\n    -   claude-code-swarm-plugin 的 plugin.json 停留在 0.59.6（package.json 为 0.60.0）\n    -   marketplace.json 停留在 0.57.5\n\n    **更新了 `sync-plugin-versions.ts`** 脚本，以同步以下三个清单文件：\n    -   opencode-swarm-plugin\u002Fclaude-plugin\u002F.claude-plugin\u002Fplugin.json\n    -   claude-code-swarm-plugin\u002F.claude-plugin\u002Fplugin.json\n    -   .claude-plugin\u002Fmarketplace.json\n\n    **在 claude-code-swarm-plugin\u002Fpackage.json 中添加了 `version` 生命周期钩子**，指向共享的同步脚本，这样当 changesets 升级任一包时，都会触发完整的版本同步。\n\n    > “微服务之所以能够实现，得益于容器化的便捷性以及计算资源的按需调配，从而简化了托管、扩展和管理流程。” —《构建事件驱动的微服务》","2026-02-06T02:00:11",{"id":210,"version":211,"summary_zh":212,"released_at":213},197604,"swarm-mail@1.11.2","### Patch Changes\n\n-   [`765e442`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F765e442407ac9d9905481460bc57192db69e4283) Thanks [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - fix(memory): self-heal missing columns in memories table\n\n    The migration system was importing PGlite migrations (`memoryMigrations`) instead\n    of libSQL migrations (`memoryMigrationsLibSQL`), causing schema drift. Columns\n    defined in the Drizzle schema (`tags`, `updated_at`, `decay_factor`, `access_count`,\n    `last_accessed`, `category`, `status`) were never added via migrations.\n\n    Added `healMemorySchema()` that runs after every migration pass — checks\n    `pragma_table_info` for missing columns and adds them idempotently. Databases\n    created via migrations, convenience functions, or PGlite migration all converge\n    on the correct schema.\n\n    Also added v12 migration marker and fixed the import to use `memoryMigrationsLibSQL`.\n","2026-02-06T02:00:08",{"id":215,"version":216,"summary_zh":217,"released_at":218},197605,"opencode-swarm-plugin@0.62.2","### Patch Changes\n\n-   [`453fe90`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F453fe9063070a483dc64bfaabbf99362c8674e48) Thanks [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - fix(versions): sync all plugin.json manifests via changesets lifecycle hook\n\n    plugin.json files were never updated by changesets, causing version drift:\n\n    -   opencode-swarm-plugin plugin.json stuck at 0.59.5 (package.json: 0.62.0)\n    -   claude-code-swarm-plugin plugin.json stuck at 0.59.6 (package.json: 0.60.0)\n    -   marketplace.json stuck at 0.57.5\n\n    **Updated `sync-plugin-versions.ts`** to sync all three manifests:\n\n    -   opencode-swarm-plugin\u002Fclaude-plugin\u002F.claude-plugin\u002Fplugin.json\n    -   claude-code-swarm-plugin\u002F.claude-plugin\u002Fplugin.json\n    -   .claude-plugin\u002Fmarketplace.json\n\n    **Added `version` lifecycle hook** to claude-code-swarm-plugin\u002Fpackage.json\n    pointing to the shared sync script so changesets bumping either package\n    triggers a full sync.\n\n    > \"Microservices are facilitated by the ease of containerization and the\n    > requisitioning of compute resources, allowing for simplified hosting,\n    > scaling, and management.\" — Building Event-Driven Microservices\n\n-   Updated dependencies \\[[`765e442`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F765e442407ac9d9905481460bc57192db69e4283)]:\n    -   swarm-mail@1.11.2\n","2026-02-06T02:00:05",{"id":220,"version":221,"summary_zh":222,"released_at":223},197606,"opencode-swarm-plugin@0.62.1","### Patch Changes\n\n-   [`552ca1a`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F552ca1a4a077bd4f61b0f3568ccf82d01d27bc13) Thanks [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - fix(hooks): restore swarm claude subcommand tree deleted by 86fab13\n\n    Commit 86fab13 (\"support multiple OpenCode installation methods\") accidentally\n    deleted ~2,070 lines from bin\u002Fswarm.ts via a bad rebase, nuking the entire\n    `swarm claude` subcommand tree. Every Claude Code hook invocation has been\n    hitting \"Unknown subcommand\" since, most visibly `agent-stop` on every response.\n\n    **Restored from 70d47d5:**\n\n    -   `case \"claude\"` in main CLI switch\n    -   ClaudeHookInput interface + 3 helper functions (readHookInput,\n        resolveClaudeProjectPath, writeClaudeHookOutput)\n    -   10 handler functions: session-start, user-prompt, pre-compact, session-end,\n        pre-edit, pre-complete, post-complete, track-tool, compliance, skill-reload\n    -   Claude admin commands: path, install, uninstall, init\n    -   Required imports: createMemoryAdapter, invalidateSkillsCache, discoverSkills\n\n    **New stub handlers for ed31f5c hooks:**\n\n    -   coordinator-start, worker-start (SubagentStart)\n    -   subagent-stop (SubagentStop), agent-stop (Stop)\n    -   track-task (PreToolUse:TaskCreate|TaskUpdate)\n    -   post-task-update (PostToolUse:TaskUpdate)\n\n    **Synced hooks.json** in opencode-swarm-plugin\u002Fclaude-plugin to include\n    SubagentStart, SubagentStop, Stop, and task tracking hooks matching\n    the claude-code-swarm-plugin version.\n\n    > \"Design fragility: the tendency of software to break in multiple places\n    > when a single change is made, often in seemingly unrelated areas.\"\n","2026-02-06T00:52:03",{"id":225,"version":226,"summary_zh":227,"released_at":228},197607,"@swarmtools\u002Fevals@0.2.36","### Patch Changes\n\n-   Updated dependencies \\[[`552ca1a`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F552ca1a4a077bd4f61b0f3568ccf82d01d27bc13)]:\n    -   opencode-swarm-plugin@0.62.1\n","2026-02-06T00:52:00",{"id":230,"version":231,"summary_zh":222,"released_at":232},197608,"claude-code-swarm-plugin@0.60.1","2026-02-06T00:51:57",{"id":234,"version":235,"summary_zh":236,"released_at":237},197609,"@swarmtools\u002Fevals@0.2.35","### Patch Changes\n\n-   Updated dependencies \\[[`ed31f5c`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002Fed31f5c316e1bb9137bb27e824f2fc58b9ba9d46), [`70d47d5`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F70d47d5c2ae3a9419a00a165e49a9f7f9d98b7d7), [`70d47d5`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F70d47d5c2ae3a9419a00a165e49a9f7f9d98b7d7)]:\n    -   opencode-swarm-plugin@0.62.0\n","2026-02-05T21:23:37",{"id":239,"version":240,"summary_zh":241,"released_at":242},197610,"opencode-swarm-plugin@0.62.0","### Minor Changes\n\n-   [#154](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fpull\u002F154) [`70d47d5`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F70d47d5c2ae3a9419a00a165e49a9f7f9d98b7d7) Thanks [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - feat: UserPromptSubmit hook now injects timestamp and semantic memory recall\n\n    -   **Timestamp injection**: Every prompt now includes current date\u002Ftime for temporal awareness\n    -   **Semantic memory recall**: Automatically searches hivemind for relevant memories on each prompt\n        -   Queries with prompt text, returns top 3 matches\n        -   Filters to high-confidence matches (score > 0.5)\n        -   Injects up to 2 relevant memory snippets as context\n    -   Uses local memory adapter wrapper for proper db type conversion\n\n-   [#154](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fpull\u002F154) [`70d47d5`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F70d47d5c2ae3a9419a00a165e49a9f7f9d98b7d7) Thanks [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - Add Pi-inspired agent enhancements:\n\n    -   **skills_reload tool**: Hot-reload skills mid-session with cache invalidation\n    -   **swarm_branch\u002Fswarm_return**: Session branching for side-quests with context forking\n    -   **skill-generator meta-skill**: Generate properly formatted skill scaffolds\n    -   **PostToolUse hook**: Auto-trigger skill reload on skills_create\u002Fupdate\u002Finit\n\n### Patch Changes\n\n-   [#163](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fpull\u002F163) [`ed31f5c`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002Fed31f5c316e1bb9137bb27e824f2fc58b9ba9d46) Thanks [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - feat(plugin): upgrade for Claude Code 2.1.32 native integration\n\n    Add dual-mode architecture supporting both native agent teams and task\n    fallback. Plugin now complements rather than duplicates native features.\n\n    **claude-code-swarm-plugin:**\n\n    -   agents: Add permissionMode, memory, disallowedTools, lifecycle hooks\n    -   swarm.md: Full rewrite with environment detection, mode-aware protocols\n    -   hooks: Add SubagentStart\u002FStop, TaskCreate\u002FTaskUpdate tracking\n    -   skills: Update for TaskCreate\u002FTaskUpdate, TeammateTool awareness\n    -   README: Add 2.1.32 integration docs, architecture diagram, comparison table\n\n    **opencode-swarm-plugin:**\n\n    -   Fix test schema mismatch: add access_count, last_accessed, category, status\n    -   Fix decay_factor default from 0.7 to 1.0 to match Drizzle schema\n    -   Update column count assertions (14 → 18 columns)\n\n    Native teams provide: real-time messaging, planning mode, task UI\n    Plugin provides: git-backed persistence, semantic memory, file locking\n\n    > \"Make the change easy, then make the easy change.\" — Kent Beck\n","2026-02-05T21:23:34",{"id":244,"version":245,"summary_zh":246,"released_at":247},197611,"claude-code-swarm-plugin@0.60.0","### Minor Changes\n\n-   [#163](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fpull\u002F163) [`ed31f5c`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002Fed31f5c316e1bb9137bb27e824f2fc58b9ba9d46) Thanks [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - feat(plugin): upgrade for Claude Code 2.1.32 native integration\n\n    Add dual-mode architecture supporting both native agent teams and task\n    fallback. Plugin now complements rather than duplicates native features.\n\n    **claude-code-swarm-plugin:**\n\n    -   agents: Add permissionMode, memory, disallowedTools, lifecycle hooks\n    -   swarm.md: Full rewrite with environment detection, mode-aware protocols\n    -   hooks: Add SubagentStart\u002FStop, TaskCreate\u002FTaskUpdate tracking\n    -   skills: Update for TaskCreate\u002FTaskUpdate, TeammateTool awareness\n    -   README: Add 2.1.32 integration docs, architecture diagram, comparison table\n\n    **opencode-swarm-plugin:**\n\n    -   Fix test schema mismatch: add access_count, last_accessed, category, status\n    -   Fix decay_factor default from 0.7 to 1.0 to match Drizzle schema\n    -   Update column count assertions (14 → 18 columns)\n\n    Native teams provide: real-time messaging, planning mode, task UI\n    Plugin provides: git-backed persistence, semantic memory, file locking\n\n    > \"Make the change easy, then make the easy change.\" — Kent Beck\n\n### Patch Changes\n\n-   [`3bbf31d`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F3bbf31d73874d49c319f4b89f51934ae9049622d) Thanks [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - fix(mcp): inline tool schemas to fix params arriving as undefined\n\n    The MCP server scraped tool definitions from `swarm tool --list --json` at\n    startup, but the CLI's `--list` handler never supported `--json`. The fallback\n    parsed colored text output and registered every tool with an empty JSON schema\n    (`properties: {}`), which converted to a Zod schema with no required fields.\n    The MCP SDK then treated all params as optional, delivering `undefined` to\n    every handler.\n\n    -   **claude-code-swarm-plugin**: Replace runtime CLI scraping with static\n        `TOOL_DEFINITIONS` array containing all 25 tools with proper JSON schemas\n        (properties, required fields, types, descriptions)\n    -   **swarm-tools**: Export `SWARM_TOOLS` from index.ts; MCP server imports\n        canonical definitions instead of scraping CLI\n    -   Remove dead `getToolDefinitions()`, `filterTools()`, unused `execSync` import\n\n    > \"The most fundamental problem in computer science is problem decomposition:\n    > how to take a complex problem and divide it up into pieces that can be solved\n    > independently.\" — John Ousterhout, A Philosophy of Software Design\n","2026-02-05T21:23:31",{"id":249,"version":250,"summary_zh":251,"released_at":252},197612,"opencode-swarm-plugin@0.61.0","### Minor Changes\n\n-   [#151](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fpull\u002F151) [`e686008`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002Fe686008d437752443af1ccca01cb2047912cf517) Thanks [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - feat: UserPromptSubmit hook now injects timestamp and semantic memory recall\n\n    -   **Timestamp injection**: Every prompt now includes current date\u002Ftime for temporal awareness\n    -   **Semantic memory recall**: Automatically searches hivemind for relevant memories on each prompt\n        -   Queries with prompt text, returns top 3 matches\n        -   Filters to high-confidence matches (score > 0.5)\n        -   Injects up to 2 relevant memory snippets as context\n    -   Uses local memory adapter wrapper for proper db type conversion\n\n### Patch Changes\n\n-   [`109f335`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F109f335b663be6420bfd8a471118dc283c5248c2) Thanks [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - Add SKOS taxonomy extraction to hivemind memory system\n\n    -   SKOS entity taxonomy with broader\u002Fnarrower\u002Frelated relationships\n    -   LLM-powered taxonomy extraction wired into adapter.store()\n    -   Entity extraction now includes prefLabel and altLabels\n    -   New CLI commands: `swarm memory entities`, `swarm memory entity`, `swarm memory taxonomy`\n    -   Moltbot plugin: decay tier filtering, entity-aware auto-capture\n    -   HATEOAS-style hints in hivemind tool responses\n\n-   Updated dependencies \\[[`109f335`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F109f335b663be6420bfd8a471118dc283c5248c2)]:\n    -   swarm-mail@1.11.1\n","2026-02-01T16:18:05",{"id":254,"version":255,"summary_zh":256,"released_at":257},197613,"swarm-mail@1.11.1","### Patch Changes\n\n-   [`109f335`](https:\u002F\u002Fgithub.com\u002Fjoelhooks\u002Fswarm-tools\u002Fcommit\u002F109f335b663be6420bfd8a471118dc283c5248c2) Thanks [@joelhooks](https:\u002F\u002Fgithub.com\u002Fjoelhooks)! - Add SKOS taxonomy extraction to hivemind memory system\n\n    -   SKOS entity taxonomy with broader\u002Fnarrower\u002Frelated relationships\n    -   LLM-powered taxonomy extraction wired into adapter.store()\n    -   Entity extraction now includes prefLabel and altLabels\n    -   New CLI commands: `swarm memory entities`, `swarm memory entity`, `swarm memory taxonomy`\n    -   Moltbot plugin: decay tier filtering, entity-aware auto-capture\n    -   HATEOAS-style hints in hivemind tool responses\n","2026-02-01T16:18:02"]