[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-covibes--zeroshot":3,"tool-covibes--zeroshot":61},[4,18,26,36,44,52],{"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",141543,2,"2026-04-06T11:32:54",[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。",107888,"2026-04-06T11:32:50",[14,15,13],{"id":45,"name":46,"github_repo":47,"description_zh":48,"stars":49,"difficulty_score":10,"last_commit_at":50,"category_tags":51,"status":17},4487,"LLMs-from-scratch","rasbt\u002FLLMs-from-scratch","LLMs-from-scratch 是一个基于 PyTorch 的开源教育项目，旨在引导用户从零开始一步步构建一个类似 ChatGPT 的大型语言模型（LLM）。它不仅是同名技术著作的官方代码库，更提供了一套完整的实践方案，涵盖模型开发、预训练及微调的全过程。\n\n该项目主要解决了大模型领域“黑盒化”的学习痛点。许多开发者虽能调用现成模型，却难以深入理解其内部架构与训练机制。通过亲手编写每一行核心代码，用户能够透彻掌握 Transformer 架构、注意力机制等关键原理，从而真正理解大模型是如何“思考”的。此外，项目还包含了加载大型预训练权重进行微调的代码，帮助用户将理论知识延伸至实际应用。\n\nLLMs-from-scratch 特别适合希望深入底层原理的 AI 开发者、研究人员以及计算机专业的学生。对于不满足于仅使用 API，而是渴望探究模型构建细节的技术人员而言，这是极佳的学习资源。其独特的技术亮点在于“循序渐进”的教学设计：将复杂的系统工程拆解为清晰的步骤，配合详细的图表与示例，让构建一个虽小但功能完备的大模型变得触手可及。无论你是想夯实理论基础，还是为未来研发更大规模的模型做准备",90106,"2026-04-06T11:19:32",[35,15,13,14],{"id":53,"name":54,"github_repo":55,"description_zh":56,"stars":57,"difficulty_score":10,"last_commit_at":58,"category_tags":59,"status":17},4292,"Deep-Live-Cam","hacksider\u002FDeep-Live-Cam","Deep-Live-Cam 是一款专注于实时换脸与视频生成的开源工具，用户仅需一张静态照片，即可通过“一键操作”实现摄像头画面的即时变脸或制作深度伪造视频。它有效解决了传统换脸技术流程繁琐、对硬件配置要求极高以及难以实时预览的痛点，让高质量的数字内容创作变得触手可及。\n\n这款工具不仅适合开发者和技术研究人员探索算法边界，更因其极简的操作逻辑（仅需三步：选脸、选摄像头、启动），广泛适用于普通用户、内容创作者、设计师及直播主播。无论是为了动画角色定制、服装展示模特替换，还是制作趣味短视频和直播互动，Deep-Live-Cam 都能提供流畅的支持。\n\n其核心技术亮点在于强大的实时处理能力，支持口型遮罩（Mouth Mask）以保留使用者原始的嘴部动作，确保表情自然精准；同时具备“人脸映射”功能，可同时对画面中的多个主体应用不同面孔。此外，项目内置了严格的内容安全过滤机制，自动拦截涉及裸露、暴力等不当素材，并倡导用户在获得授权及明确标注的前提下合规使用，体现了技术发展与伦理责任的平衡。",88924,"2026-04-06T03:28:53",[14,15,13,60],"视频",{"id":62,"github_repo":63,"name":64,"description_en":65,"description_zh":66,"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":77,"owner_email":77,"owner_twitter":77,"owner_website":78,"owner_url":79,"languages":80,"stars":104,"forks":105,"last_commit_at":106,"license":107,"difficulty_score":32,"env_os":108,"env_gpu":109,"env_ram":109,"env_deps":110,"category_tags":121,"github_topics":123,"view_count":32,"oss_zip_url":77,"oss_zip_packed_at":77,"status":17,"created_at":142,"updated_at":143,"faqs":144,"releases":172},4406,"covibes\u002Fzeroshot","zeroshot","Your autonomous engineering team in a CLI. Point Zeroshot at an issue, walk away, and return to production-grade code. Supports Claude Code, OpenAI Codex, OpenCode, and Gemini CLI.","zeroshot 是一款开源的命令行工具，旨在打造你的“自主工程团队”。它不仅仅是一个代码生成器，而是一个能够独立完成任务规划、代码实现、测试验证及迭代修复的多智能体编排系统。用户只需指向一个任务（如 GitHub Issue、本地文件或一段文字描述），zeroshot 便会自动在隔离环境中运行，直到产出经过严格验证的生产级代码，或在失败时提供可复现的具体原因。\n\n传统 AI 编程助手往往缺乏自我验证能力，容易陷入“盲目自信”的误区。zeroshot 通过引入独立的“规划者”、“执行者”和“验证者”角色，解决了这一痛点。它内置了盲测机制，会循环执行“计划 - 实施 - 验证”流程，确保代码不仅被写出，更能通过自动化测试并符合验收标准，特别适用于对正确性要求高于速度的复杂场景。\n\n这款工具非常适合注重代码质量与可复现性的高级工程师、希望自动化 PR 工作流的开发团队，以及需要处理大量遗留问题的开源维护者。其独特亮点在于支持多种主流 AI 模型（如 Claude Code、OpenAI Codex、Gemini 等）作为后端，并兼容 GitHub、GitLab、Jira 等多个任务管理平台","zeroshot 是一款开源的命令行工具，旨在打造你的“自主工程团队”。它不仅仅是一个代码生成器，而是一个能够独立完成任务规划、代码实现、测试验证及迭代修复的多智能体编排系统。用户只需指向一个任务（如 GitHub Issue、本地文件或一段文字描述），zeroshot 便会自动在隔离环境中运行，直到产出经过严格验证的生产级代码，或在失败时提供可复现的具体原因。\n\n传统 AI 编程助手往往缺乏自我验证能力，容易陷入“盲目自信”的误区。zeroshot 通过引入独立的“规划者”、“执行者”和“验证者”角色，解决了这一痛点。它内置了盲测机制，会循环执行“计划 - 实施 - 验证”流程，确保代码不仅被写出，更能通过自动化测试并符合验收标准，特别适用于对正确性要求高于速度的复杂场景。\n\n这款工具非常适合注重代码质量与可复现性的高级工程师、希望自动化 PR 工作流的开发团队，以及需要处理大量遗留问题的开源维护者。其独特亮点在于支持多种主流 AI 模型（如 Claude Code、OpenAI Codex、Gemini 等）作为后端，并兼容 GitHub、GitLab、Jira 等多个任务管理平台。此外，它还具备崩溃恢复功能，能持久化保存状态，确保长耗时任务的安全执行。","# zeroshot CLI\n\n> **🎉 New in v5.4:** Now supports **OpenCode** CLI! Use Claude, Codex, Gemini, or OpenCode as your AI provider. Also supports **GitHub, GitLab, Jira, and Azure DevOps** as issue backends. See [Providers](#providers) and [Multi-Platform Issue Support](#multi-platform-issue-support).\n\n\u003C!-- install-placeholder -->\n\u003Cp align=\"center\">\n \u003Ccode>npm install -g @covibes\u002Fzeroshot\u003C\u002Fcode>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcovibes_zeroshot_readme_bb271f400df2.gif\" alt=\"Demo\" width=\"700\">\n \u003Cbr>\n \u003Cem>Demo (100x speed, 90-minute run, 5 iterations to approval)\u003C\u002Fem>\n\u003C\u002Fp>\n\n[![CI](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Factions\u002Fworkflows\u002Fci.yml\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Factions\u002Fworkflows\u002Fci.yml)\n[![npm version](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fv\u002F@covibes\u002Fzeroshot.svg)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@covibes\u002Fzeroshot)\n[![License: MIT](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-yellow.svg)](LICENSE)\n[![Node 18+](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fnode-18%2B-brightgreen.svg)](https:\u002F\u002Fnodejs.org\u002F)\n![Platform: Linux | macOS](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fplatform-Linux%20%7C%20macOS-blue.svg)\n\n\u003C!-- discord-placeholder -->\n\n[![Discord](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDiscord-Join-5865F2?logo=discord&logoColor=white)](https:\u002F\u002Fdiscord.gg\u002FPdZ3UEXB)\n\nZeroshot is an open-source AI coding agent orchestration CLI that runs multi-agent workflows to autonomously implement, review, test, and verify code changes.\n\nIt runs a **planner**, an **implementer**, and independent **validators** in isolated environments, looping until changes are **verified** or **rejected** with actionable, reproducible failures.\n\nBuilt for tasks where correctness matters more than speed.\n\n## How It Works\n\n- Plan: translate a task into concrete acceptance criteria\n- Implement: make changes in an isolated workspace (local, worktree, or Docker)\n- Validate: run automated checks with independent validators\n- Iterate: repeat until verified, or return actionable failures\n- Resume: crash-safe state persisted for recovery\n\n## Quick Start\n\n```bash\nzeroshot run 123 # GitHub issue number\nzeroshot run feature.md # Markdown file\nzeroshot run \"Add dark mode\" # Inline text\n```\n\nOr describe a complex task inline:\n\n```bash\nzeroshot run \"Add optimistic locking with automatic retry: when updating a user,\nretry with exponential backoff up to 3 times, merge non-conflicting field changes,\nand surface conflicts with details. Handle the ABA problem where version goes A->B->A.\"\n```\n\n## Why Not Just Use a Single AI Agent?\n\n| Approach | Writes Code | Runs Tests | Blind Validation | Iterates Until Verified |\n| -------------------------- | ----------- | ---------- | ---------------- | ----------------------- |\n| Chat-based assistant | ✅ | ⚠️ | ❌ | ❌ |\n| Single coding agent | ✅ | ⚠️ | ❌ | ⚠️ |\n| **Zeroshot (multi-agent)** | ✅ | ✅ | ✅ | ✅ |\n\n## Use Cases\n\n- Autonomous AI code refactoring\n- AI-powered pull request automation\n- Automated bug fixing with validation\n- Multi-agent code generation for software engineering\n- Agentic coding workflows with blind validation\n\n## Who Is This For?\n\n- Senior engineers who care about correctness and reproducibility\n- Teams automating PR workflows and code review gates\n- Infra\u002Fplatform teams standardizing agentic workflows\n- Open-source maintainers working through issue backlogs\n- AI power users who want verification, not vibes\n\n## Install and Requirements\n\n**Platforms**: Linux, macOS. Windows (native\u002FWSL) is deferred while we harden reliability and multi-provider correctness.\n\n```bash\nnpm install -g @covibes\u002Fzeroshot\n```\n\n**Requires**: Node 18+, at least one provider CLI (Claude Code, Codex, Gemini, Opencode).\n\n```bash\n# Install one or more providers\nnpm i -g @anthropic-ai\u002Fclaude-code\nnpm i -g @openai\u002Fcodex\nnpm i -g @google\u002Fgemini-cli\n# Opencode: see https:\u002F\u002Fopencode.ai\n\n# Authenticate with the provider CLI\nclaude login # Claude\ncodex login # Codex\ngemini auth login # Gemini\nopencode auth login # Opencode\n\n# GitHub auth (for issue numbers)\ngh auth login\n```\n\n## Providers\n\nZeroshot shells out to provider CLIs. Pick a default and override per run:\n\n```bash\nzeroshot providers\nzeroshot providers set-default codex\nzeroshot run 123 --provider gemini\n```\n\nSee `docs\u002Fproviders.md` for setup, model levels, and Docker mounts.\n\n## Why Multiple Agents?\n\nSingle-agent sessions degrade. Context gets buried under thousands of tokens. The model optimizes for \"done\" over \"correct.\"\n\nZeroshot fixes this with isolated agents that check each other's work. Validators can't lie about code they didn't write. Fail the check? Fix and retry until it actually works.\n\n## What Makes It Different\n\n- **Blind validation** - Validators never see the worker's context or code history\n- **Repeatable workflows** - Task complexity determines agent count and model selection\n- **Accept\u002Freject loop** - Rejections include actionable findings, not vague complaints\n- **Crash recovery** - All state persisted to SQLite; resume anytime\n- **Isolation modes** - None, git worktree, or Docker container\n- **Cost control** - Model ceilings prevent runaway API spend\n\n## When to Use Zeroshot\n\nZeroshot performs best when tasks have clear acceptance criteria.\n\n| Scenario | Use | Why |\n| ----------------------------------------------- | --- | ------------------------- |\n| Add rate limiting (sliding window, per-IP, 429) | Yes | Clear requirements |\n| Refactor auth to JWT | Yes | Defined end state |\n| Fix login bug | Yes | Success is measurable |\n| Fix 2410 lint violations | Yes | Clear completion criteria |\n| Make the app faster | No | Needs exploration first |\n| Improve the codebase | No | No acceptance criteria |\n| Figure out flaky tests | No | Exploratory |\n\nRule of thumb: if you cannot describe what \"done\" means, validators cannot verify it.\n\n## Command Overview\n\n```bash\n# Run\nzeroshot run 123 # GitHub issue\nzeroshot run feature.md # Markdown file\nzeroshot run \"Add dark mode\" # Inline text\n\n# Isolation\nzeroshot run 123 --worktree # git worktree\nzeroshot run 123 --docker # container\n\n# Automation (--ship implies --pr implies --worktree)\nzeroshot run 123 --pr # worktree + create PR\nzeroshot run 123 --ship # PR + auto-merge on approval\n\n# Background mode\nzeroshot run 123 -d\nzeroshot run 123 --ship -d\n\n# Control\nzeroshot list\nzeroshot status \u003Cid>\nzeroshot logs \u003Cid> -f\nzeroshot resume \u003Cid>\nzeroshot stop \u003Cid>\nzeroshot kill \u003Cid>\nzeroshot watch\n\n# Providers\nzeroshot providers\nzeroshot providers set-default codex\n\n# Agent library\nzeroshot agents list\nzeroshot agents show \u003Cname>\n\n# Maintenance\nzeroshot clean\nzeroshot purge\n```\n\n## Multi-Platform Issue Support\n\nZeroshot works with **GitHub, GitLab, Jira, and Azure DevOps**. Just paste the issue URL or key.\nWhen working in a git repository, zeroshot automatically detects the issue provider from your git remote URL. No configuration needed!\n\n```bash\n# GitHub\nzeroshot run 123\nzeroshot run https:\u002F\u002Fgithub.com\u002Forg\u002Frepo\u002Fissues\u002F123\n\n# GitLab (cloud and self-hosted)\nzeroshot run https:\u002F\u002Fgitlab.com\u002Forg\u002Frepo\u002F-\u002Fissues\u002F456\nzeroshot run https:\u002F\u002Fgitlab.mycompany.com\u002Forg\u002Frepo\u002F-\u002Fissues\u002F789\n\n# Jira\nzeroshot run PROJ-789\nzeroshot run https:\u002F\u002Fcompany.atlassian.net\u002Fbrowse\u002FPROJ-789\n\n# Azure DevOps\nzeroshot run https:\u002F\u002Fdev.azure.com\u002Forg\u002Fproject\u002F_workitems\u002Fedit\u002F999\n```\n\n**Requires**: CLI tools ([`gh`](https:\u002F\u002Fcli.github.com\u002F), [`glab`](https:\u002F\u002Fgitlab.com\u002Fgitlab-org\u002Fcli), [`jira`](https:\u002F\u002Fgithub.com\u002Fgo-jira\u002Fjira), or [`az`](https:\u002F\u002Fdocs.microsoft.com\u002Fcli\u002Fazure\u002F)) for the platform you use. See [issue-providers README](src\u002Fissue-providers\u002FREADME.md) for setup and self-hosted instances.\n\n**Important for `--pr` mode**: Run zeroshot from the target repository directory. PRs are created on the git remote of your current directory. If you run from a different repo, zeroshot will warn you and skip the \"Closes #X\" reference (the PR is still created, but won't auto-close the issue).\n\n## Architecture\n\nZeroshot is a message-driven coordination layer with smart defaults.\n\n- The conductor classifies tasks by complexity and type.\n- A workflow template selects agents and validators.\n- Agents publish results to a SQLite ledger.\n- Validators approve or reject with specific findings.\n- Rejections route back to the worker for fixes.\n\n```\n ┌─────────────────┐\n │ TASK │\n └────────┬────────┘\n │\n ▼\n ┌────────────────────────────────────────────┐\n │ CONDUCTOR │\n │ Complexity × TaskType → Workflow │\n └────────────────────────┬───────────────────┘\n │\n ┌─────────────────────────────┼─────────────────────────────┐\n │ │ │\n ▼ ▼ ▼\n ┌───────────┐ ┌───────────┐ ┌───────────┐\n │ TRIVIAL │ │ SIMPLE │ │ STANDARD+ │\n │ 1 agent │──────────▶ │ worker │ │ planner │\n │ (level1) │ COMPLETE │ + 1 valid.│ │ + worker │\n │ no valid. │ └─────┬─────┘ │ + 3-5 val.│\n └───────────┘ │ └─────┬─────┘\n ▼ │\n ┌─────────────┐ ▼\n ┌──▶│ WORKER │ ┌─────────────┐\n │ └──────┬──────┘ │ PLANNER │\n │ │ └──────┬──────┘\n │ ▼ │\n │ ┌─────────────────────┐ ▼\n │ │ ✓ validator │ ┌─────────────┐\n │ │ (generic check) │ ┌──▶│ WORKER │\n │ └──────────┬──────────┘ │ └──────┬──────┘\n │ REJECT │ ALL OK │ │\n └──────────────┘ │ │ ▼\n │ │ ┌──────────────────────┐\n │ │ │ ✓ requirements │\n │ │ │ ✓ code (STANDARD+) │\n │ │ │ ✓ security (CRIT) │\n │ │ │ ✓ tester (CRIT) │\n │ │ │ ✓ adversarial │\n │ │ │ (real execution) │\n │ │ └──────────┬───────────┘\n │ │ REJECT │ ALL OK\n │ └──────────────┘ │\n ▼ ▼\n ┌─────────────────────────────────────────────────────────────────────────────┐\n │ COMPLETE │\n └─────────────────────────────────────────────────────────────────────────────┘\n```\n\n### Complexity Model\n\n| Task | Complexity | Agents | Validators |\n| ---------------------- | ---------- | ------ | ------------------------------------------------- |\n| Fix typo in README | TRIVIAL | 1 | None |\n| Add dark mode toggle | SIMPLE | 2 | Generic validator |\n| Refactor auth system | STANDARD | 4 | Requirements, code |\n| Implement payment flow | CRITICAL | 7 | Requirements, code, security, tester, adversarial |\n\n### Model Selection by Complexity\n\n| Complexity | Planner | Worker | Validators |\n| ---------- | ------- | ------ | ---------- |\n| TRIVIAL | - | level1 | - |\n| SIMPLE | - | level2 | 1 (level2) |\n| STANDARD | level2 | level2 | 2 (level2) |\n| CRITICAL | level3 | level2 | 5 (level2) |\n\nLevels map to provider-specific models. Configure with `zeroshot providers setup \u003Cprovider>` or\n`settings.providerSettings`. (Legacy `maxModel` applies to Claude only.)\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Custom Workflows (Framework Mode)\u003C\u002Fstrong>\u003C\u002Fsummary>\n\nZeroshot is message-driven, so you can define any agent topology.\n\n- Expert panels: parallel specialists -> aggregator -> decision\n- Staged gates: sequential validators, each with veto power\n- Hierarchical: supervisor dynamically spawns workers\n- Dynamic: conductor adds agents mid-execution\n\n**Coordination primitives:**\n\n- Message bus (pub\u002Fsub topics)\n- Triggers (wake agents on conditions)\n- Ledger (SQLite, crash recovery)\n- Dynamic spawning (CLUSTER_OPERATIONS)\n\n#### Creating Custom Clusters with a Provider CLI\n\nStart your provider CLI and describe your cluster:\n\n```\nCreate a zeroshot cluster config for security-critical features:\n\n1. Implementation agent (level2) implements the feature\n2. FOUR parallel validators:\n - Security validator: OWASP checks, SQL injection, XSS, CSRF\n - Performance validator: No N+1 queries, proper indexing\n - Privacy validator: GDPR compliance, data minimization\n - Code reviewer: General code quality\n\n3. ALL validators must approve before merge\n4. If ANY validator rejects, implementation agent fixes and resubmits\n5. Use level3 for security validator (highest stakes)\n\nLook at cluster-templates\u002Fbase-templates\u002Ffull-workflow.json\nand create a similar cluster. Save to cluster-templates\u002Fsecurity-review.json\n```\n\nBuilt-in validation checks for missing triggers, deadlocks, and invalid type wiring before running.\n\nSee [CLAUDE.md](.\u002FCLAUDE.md) for the cluster schema and examples.\n\n\u003C\u002Fdetails>\n\n## Crash Recovery\n\nAll state is persisted in the SQLite ledger. You can resume at any time:\n\n```bash\nzeroshot resume cluster-bold-panther\n```\n\n## Isolation Modes\n\n### Git Worktree (Default for --pr\u002F--ship)\n\n```bash\nzeroshot run 123 --worktree\n```\n\nLightweight isolation using git worktree. Creates a separate working directory with its own branch. Auto-enabled with `--pr` and `--ship`.\n\n### Docker Container\n\n```bash\nzeroshot run 123 --docker\n```\n\nFull isolation in a fresh container. Your workspace stays untouched. Useful for risky experiments or parallel runs.\n\n### When to Use Which\n\n| Scenario | Recommended |\n| ------------------------------------ | ---------------------- |\n| Quick task, review changes yourself | No isolation (default) |\n| PR workflow, code review | `--worktree` or `--pr` |\n| Risky experiment, might break things | `--docker` |\n| Running multiple tasks in parallel | `--docker` |\n| Full automation, no review needed | `--ship` |\n\n**Default behavior:** Agents modify files only; they do not commit or push unless using an isolation mode that explicitly allows it.\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Docker Credential Mounts\u003C\u002Fstrong>\u003C\u002Fsummary>\n\nWhen using `--docker`, zeroshot mounts credential directories so agents can access provider CLIs and tools like AWS, Azure, and kubectl.\n\n**Default mounts**: `gh`, `git`, `ssh` (GitHub CLI, git config, SSH keys)\n\n**Available presets**: `gh`, `git`, `ssh`, `aws`, `azure`, `kube`, `terraform`, `gcloud`, `claude`, `codex`, `gemini`\n\n```bash\n# Configure via settings (persistent)\nzeroshot settings set dockerMounts '[\"gh\", \"git\", \"ssh\", \"aws\", \"azure\"]'\n\n# View current config\nzeroshot settings get dockerMounts\n\n# Per-run override\nzeroshot run 123 --docker --mount ~\u002F.aws:\u002Froot\u002F.aws:ro\n\n# Provider credentials\nzeroshot run 123 --docker --mount ~\u002F.config\u002Fcodex:\u002Fhome\u002Fnode\u002F.config\u002Fcodex:ro\nzeroshot run 123 --docker --mount ~\u002F.config\u002Fgemini:\u002Fhome\u002Fnode\u002F.config\u002Fgemini:ro\n\n# Disable all mounts\nzeroshot run 123 --docker --no-mounts\n\n# CI: env var override\nZEROSHOT_DOCKER_MOUNTS='[\"aws\",\"azure\"]' zeroshot run 123 --docker\n```\n\nSee `docs\u002Fproviders.md` for provider CLI setup and mount details.\n\n**Custom mounts** (mix presets with explicit paths):\n\n```bash\nzeroshot settings set dockerMounts '[\n \"gh\",\n \"git\",\n {\"host\": \"~\u002F.myconfig\", \"container\": \"$HOME\u002F.myconfig\", \"readonly\": true}\n]'\n```\n\n**Container home**: Presets use `$HOME` placeholder. Default: `\u002Froot`. Override with:\n\n```bash\nzeroshot settings set dockerContainerHome '\u002Fhome\u002Fnode'\n# Or per-run:\nzeroshot run 123 --docker --container-home \u002Fhome\u002Fnode\n```\n\n**Env var passthrough**: Presets auto-pass related env vars (for example, `aws` -> `AWS_REGION`, `AWS_PROFILE`). Add custom:\n\n```bash\nzeroshot settings set dockerEnvPassthrough '[\"MY_API_KEY\", \"TF_VAR_*\"]'\n```\n\n\u003C\u002Fdetails>\n\n## Resources\n\n- [CLAUDE.md](.\u002FCLAUDE.md) - Architecture, cluster config schema, agent primitives\n- `docs\u002Fproviders.md` - Provider setup, model levels, and Docker mounts\n- `docs\u002Fcontext-management.md` - Context selection, context packs, and state snapshots\n- [Discord](https:\u002F\u002Fdiscord.gg\u002FPdZ3UEXB) - Support and community\n- `zeroshot export \u003Cid>` - Export conversation to markdown\n- `sqlite3 ~\u002F.zeroshot\u002F*.db` - Direct ledger access for debugging\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Troubleshooting\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n| Issue | Fix |\n| ----------------------------- | ----------------------------------------------------------------------------------------- |\n| `claude: command not found` | `npm i -g @anthropic-ai\u002Fclaude-code && claude auth login` |\n| `codex: command not found` | `npm i -g @openai\u002Fcodex && codex login` |\n| `gemini: command not found` | `npm i -g @google\u002Fgemini-cli && gemini auth login` |\n| `gh: command not found` | [Install GitHub CLI](https:\u002F\u002Fcli.github.com\u002F) |\n| `--docker` fails | Docker must be running: `docker ps` to verify |\n| Cluster stuck | `zeroshot resume \u003Cid>` to continue |\n| Agent keeps failing | Check `zeroshot logs \u003Cid>` for actual error |\n| `zeroshot: command not found` | `npm install -g @covibes\u002Fzeroshot` |\n| Agents misbehave | `\u002Fanalyze-cluster-postmortem \u003Cid>` in Claude Code (creates issue if fix is generalizable) |\n\n\u003C\u002Fdetails>\n\n## Contributing\n\nSee [CONTRIBUTING.md](CONTRIBUTING.md) for development setup and guidelines.\n\nPlease read [CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md) before participating.\n\nFor security issues, see [SECURITY.md](SECURITY.md).\n\n## TUI\n\nRatatui (Rust) is the only supported TUI. Entry points:\n\n- `zeroshot` (TTY + no args)\n- `zeroshot tui`\n- `zeroshot watch`\n\n### TUI Development\n\nThe Rust TUI spawns a Node backend over stdio. Run both while iterating.\n\nSingle command dev loop (auto-rebuild + restart):\n\n```bash\ncargo install cargo-watch\nnpm run dev:tui\n```\n\n1. Install deps\n\n ```bash\n npm ci\n ```\n\n2. Build the TUI backend in watch mode\n\n ```bash\n npm run build:tui-backend -- --watch\n # or\n npx tsc -p tsconfig.tui-backend.json -w\n ```\n\n3. Run the Rust TUI (second terminal)\n\n ```bash\n cd tui-rs\n cargo run -p zeroshot-tui -- --ui disruptive\n ```\n\n### Local CLI From This Repo\n\nIf you want `zeroshot` to run from the dev branch globally:\n\n```bash\nnpm run dev:link\n```\n\nOne command to link + run the TUI dev loop:\n\n```bash\nnpm run dev:bootstrap\n```\n\n### CLI Integration Loop\n\nUse the Node CLI to launch your local Rust binary:\n\n```bash\ncd tui-rs\ncargo build -p zeroshot-tui\ncd ..\nZEROSHOT_TUI_BINARY_PATH=\"$PWD\u002Ftui-rs\u002Ftarget\u002Fdebug\u002Fzeroshot-tui\" node cli\u002Findex.js tui\n```\n\n### TUI Overrides\n\n- `ZEROSHOT_TUI_BACKEND_PATH` points to a specific `lib\u002Ftui-backend\u002Fserver.js`\n- `ZEROSHOT_TUI_BINARY_PATH` points to a local Rust binary\n- `ZEROSHOT_TUI_UI=classic|disruptive` forces UI variant\n\n---\n\nMIT - [Covibes](https:\u002F\u002Fgithub.com\u002Fcovibes)\n\nBuilt on [Claude Code](https:\u002F\u002Fclaude.com\u002Fproduct\u002Fclaude-code) by Anthropic.\n","# 零样本 CLI\n\n> **🎉 v5.4 新特性：** 现已支持 **OpenCode** CLI！您可以选择 Claude、Codex、Gemini 或 OpenCode 作为 AI 提供商。同时支持 **GitHub、GitLab、Jira 和 Azure DevOps** 作为问题后端。详情请参阅 [提供商](#providers) 和 [多平台问题支持](#multi-platform-issue-support)。\n\n\u003C!-- install-placeholder -->\n\u003Cp align=\"center\">\n \u003Ccode>npm install -g @covibes\u002Fzeroshot\u003C\u002Fcode>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcovibes_zeroshot_readme_bb271f400df2.gif\" alt=\"演示\" width=\"700\">\n \u003Cbr>\n \u003Cem>演示（100倍速，90分钟运行，5轮迭代直至批准）\u003C\u002Fem>\n\u003C\u002Fp>\n\n[![CI](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Factions\u002Fworkflows\u002Fci.yml\u002Fbadge.svg)](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Factions\u002Fworkflows\u002Fci.yml)\n[![npm 版本](https:\u002F\u002Fimg.shields.io\u002Fnpm\u002Fv\u002F@covibes\u002Fzeroshot.svg)](https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002F@covibes\u002Fzeroshot)\n[![许可证：MIT](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-yellow.svg)](LICENSE)\n[![Node 18+](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fnode-18%2B-brightgreen.svg)](https:\u002F\u002Fnodejs.org\u002F)\n![平台：Linux | macOS](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fplatform-Linux%20%7C%20macOS-blue.svg)\n\n\u003C!-- discord-placeholder -->\n\n[![Discord](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDiscord-Join-5865F2?logo=discord&logoColor=white)](https:\u002F\u002Fdiscord.gg\u002FPdZ3UEXB)\n\nZeroshot 是一个开源的 AI 编码代理编排 CLI，它运行多代理工作流，以自主方式实现、审查、测试和验证代码变更。\n\n它在隔离环境中运行一个 **规划器**、一个 **实施者** 和独立的 **验证者**，循环执行直到变更被 **验证** 或因可操作且可复现的失败而被 **拒绝**。\n\n专为那些正确性比速度更重要的任务而设计。\n\n## 工作原理\n\n- 规划：将任务转化为具体的验收标准\n- 实施：在隔离的工作区（本地、工作树或 Docker）中进行更改\n- 验证：由独立的验证者运行自动化检查\n- 迭代：重复执行，直到验证通过，或返回可操作的失败信息\n- 恢复：崩溃安全的状态会被持久化以便恢复\n\n## 快速开始\n\n```bash\nzeroshot run 123 # GitHub 问题编号\nzeroshot run feature.md # Markdown 文件\nzeroshot run \"添加暗模式\" # 内联文本\n```\n\n或者直接描述一个复杂任务：\n\n```bash\nzeroshot run \"添加带有自动重试的乐观锁：更新用户时，使用指数退避策略最多重试 3 次，合并不冲突的字段更改，并详细提示冲突。处理版本从 A 变为 B 再变回 A 的 ABA 问题。\"\n```\n\n## 为什么不能只用一个 AI 代理？\n\n| 方法 | 编写代码 | 运行测试 | 盲目验证 | 迭代直至验证 |\n| -------------------------- | ----------- | ---------- | ---------------- | ----------------------- |\n| 基于聊天的助手 | ✅ | ⚠️ | ❌ | ❌ |\n| 单一编码代理 | ✅ | ⚠️ | ❌ | ⚠️ |\n| **Zeroshot（多代理）** | ✅ | ✅ | ✅ | ✅ |\n\n## 使用场景\n\n- 自主 AI 代码重构\n- AI 驱动的拉取请求自动化\n- 带有验证的自动化 bug 修复\n- 软件工程中的多代理代码生成\n- 带有盲验证的代理式编码工作流\n\n## 适合人群\n\n- 关注正确性和可重复性的高级工程师\n- 自动化 PR 流程和代码审查关卡的团队\n- 标准化代理式工作流的基础架构\u002F平台团队\n- 处理问题积压的开源维护者\n- 希望获得验证而非“感觉”的 AI 高级用户\n\n## 安装与要求\n\n**平台**：Linux、macOS。Windows（原生\u002FWSL）暂未支持，因为我们正在强化可靠性和多提供商的正确性。\n\n```bash\nnpm install -g @covibes\u002Fzeroshot\n```\n\n**要求**：Node 18+，至少一个提供商 CLI（Claude Code、Codex、Gemini、Opencode）。\n\n```bash\n# 安装一个或多个提供商\nnpm i -g @anthropic-ai\u002Fclaude-code\nnpm i -g @openai\u002Fcodex\nnpm i -g @google\u002Fgemini-cli\n# Opencode：详见 https:\u002F\u002Fopencode.ai\n\n# 使用提供商 CLI 进行认证\nclaude login # Claude\ncodex login # Codex\ngemini auth login # Gemini\nopencode auth login # Opencode\n\n# GitHub 认证（用于问题编号）\ngh auth login\n```\n\n## 提供商\n\nZeroshot 会调用各个提供商的 CLI。您可以选择默认提供商，也可以在每次运行时覆盖：\n\n```bash\nzeroshot providers\nzeroshot providers set-default codex\nzeroshot run 123 --provider gemini\n```\n\n有关设置、模型级别和 Docker 挂载，请参阅 `docs\u002Fproviders.md`。\n\n## 为什么需要多个代理？\n\n单代理会话容易失效。上下文会在数千个 token 中被淹没。模型更倾向于“完成”而不是“正确”。\n\nZeroshot 通过隔离的代理来解决这个问题，这些代理会相互检查彼此的工作。验证者无法对未编写的代码撒谎。如果检查失败？就修复并重试，直到真正有效为止。\n\n## 区别于其他工具之处\n\n- **盲验证**：验证者永远不会看到工作者的上下文或代码历史\n- **可重复的工作流**：任务复杂度决定了代理数量和模型选择\n- **接受\u002F拒绝循环**：拒绝结果包含可操作的发现，而非含糊的抱怨\n- **崩溃恢复**：所有状态都会持久化到 SQLite 数据库中；随时可以恢复\n- **隔离模式**：无隔离、git 工作树或 Docker 容器\n- **成本控制**：模型上限可防止 API 费用失控\n\n## 何时使用 Zeroshot\n\nZeroshot 在任务具有明确的验收标准时表现最佳。\n\n| 场景 | 使用 | 为什么 |\n| ----------------------------------------------- | --- | ------------------------- |\n| 添加限流（滑动窗口，按 IP 限制，429 错误） | 是 | 要求清晰 |\n| 将认证重构为 JWT | 是 | 定义了最终状态 |\n| 修复登录 bug | 是 | 成功与否可衡量 |\n| 修复 2410 个 Lint 违规 | 是 | 完成标准明确 |\n| 让应用更快 | 否 | 需要先探索 |\n| 改善代码库 | 否 | 无验收标准 |\n| 查明不稳定测试 | 否 | 探索性任务 |\n\n经验法则：如果您无法描述“完成”意味着什么，那么验证者就无法对其进行验证。\n\n## 命令概览\n\n```bash\n# 运行\nzeroshot run 123 # GitHub 问题\nzeroshot run feature.md # Markdown 文件\nzeroshot run \"添加暗模式\" # 内联文本\n\n# 隔离\nzeroshot run 123 --worktree # git 工作树\nzeroshot run 123 --docker # 容器\n\n# 自动化（--ship 表示 --pr 表示 --worktree）\nzeroshot run 123 --pr # 工作树 + 创建 PR\nzeroshot run 123 --ship # PR + 批准时自动合并\n\n# 后台模式\nzeroshot run 123 -d\nzeroshot run 123 --ship -d\n\n# 控制\nzeroshot list\nzeroshot status \u003Cid>\nzeroshot logs \u003Cid> -f\nzeroshot resume \u003Cid>\nzeroshot stop \u003Cid>\nzeroshot kill \u003Cid>\nzeroshot watch\n\n# 提供商\nzeroshot 提供商\nzeroshot 提供商 设置默认 codex\n\n# 代理库\nzeroshot 代理 列表\nzeroshot 代理 显示 \u003C名称>\n\n# 维护\nzeroshot 清理\nzeroshot 清除\n```\n\n## 多平台问题支持\n\nZeroshot 支持 **GitHub、GitLab、Jira 和 Azure DevOps**。只需粘贴问题的 URL 或编号即可。\n在 Git 仓库中工作时，zeroshot 会自动从你的 Git 远程 URL 中检测问题提供商，无需任何配置！\n\n```bash\n# GitHub\nzeroshot 运行 123\nzeroshot 运行 https:\u002F\u002Fgithub.com\u002Forg\u002Frepo\u002Fissues\u002F123\n\n# GitLab（云和自托管）\nzeroshot 运行 https:\u002F\u002Fgitlab.com\u002Forg\u002Frepo\u002F-\u002Fissues\u002F456\nzeroshot 运行 https:\u002F\u002Fgitlab.mycompany.com\u002Forg\u002Frepo\u002F-\u002Fissues\u002F789\n\n# Jira\nzeroshot 运行 PROJ-789\nzeroshot 运行 https:\u002F\u002Fcompany.atlassian.net\u002Fbrowse\u002FPROJ-789\n\n# Azure DevOps\nzeroshot 运行 https:\u002F\u002Fdev.azure.com\u002Forg\u002Fproject\u002F_workitems\u002Fedit\u002F999\n```\n\n**要求**：根据你使用的平台，需要相应的 CLI 工具（[`gh`](https:\u002F\u002Fcli.github.com\u002F)、[`glab`](https:\u002F\u002Fgitlab.com\u002Fgitlab-org\u002Fcli\u002F)、[`jira`](https:\u002F\u002Fgithub.com\u002Fgo-jira\u002Fjira) 或 [`az`](https:\u002F\u002Fdocs.microsoft.com\u002Fcli\u002Fazure\u002F))。有关设置及自托管实例，请参阅 [issue-providers README](src\u002Fissue-providers\u002FREADME.md)。\n\n**`--pr` 模式注意事项**：请从目标仓库目录运行 zeroshot。拉取请求将在你当前目录的 Git 远程仓库上创建。如果你从其他仓库运行，zeroshot 会发出警告并跳过“Closes #X”引用（拉取请求仍会被创建，但不会自动关闭问题）。\n\n## 架构\n\nZeroshot 是一个基于消息驱动的协调层，具有智能默认配置。\n\n- 调度器会根据任务的复杂性和类型对任务进行分类。\n- 工作流模板会选择代理和验证者。\n- 代理会将结果发布到 SQLite 账本中。\n- 验证者会根据具体发现批准或拒绝。\n- 被拒绝的任务会返回给执行者进行修复。\n\n```\n ┌─────────────────┐\n │ 任务 │\n └────────┬────────┘\n │\n ▼\n ┌────────────────────────────────────────────┐\n │ 调度器 │\n │ 复杂度 × 任务类型 → 工作流 │\n └────────────────────────┬───────────────────┘\n │\n ┌─────────────────────────────┼─────────────────────────────┐\n │ │ │\n ▼ ▼ ▼\n ┌───────────┐ ┌───────────┐ ┌───────────┐\n │ 简单任务 │ │ 简单任务 │ │ 标准及以上任务 │\n │ 1 个代理 │──────────▶ │ 执行者 │ │ 计划者 │\n │ (初级) │ 完成 │ + 1 个验证者│ │ + 执行者 │\n │ 无验证者 │ └─────┬─────┘ │ + 3-5 个验证者│\n └───────────┘ │ └─────┬─────┘\n ▼ │\n ┌─────────────┐ ▼\n ┌──▶│ 执行者 │ ┌─────────────┐\n │ └──────┬──────┘ │ 计划者 │\n │ │ └──────┬──────┘\n │ ▼ │\n │ ┌─────────────────────┐ ▼\n │ │ ✓ 验证者 │ ┌─────────────┐\n │ │ （通用检查） │ ┌──▶│ 执行者 │\n │ └──────────┬──────────┘ │ └──────┬──────┘\n │ 拒绝 │ 全部通过 │ │\n └──────────────┘ │ │ ▼\n │ │ ┌──────────────────────┐\n │ │ │ ✓ 要求 │\n │ │ │ ✓ 代码（标准及以上） │\n │ │ │ ✓ 安全性（关键） │\n │ │ │ ✓ 测试员（关键） │\n │ │ │ ✓ 对抗测试 │\n │ │ │ （实际执行） │\n │ │ │ （实际执行） │\n │ │ └──────────┬───────────┘\n │ │ 拒绝 │ 全部通过\n │ └──────────────┘ │\n ▼ ▼\n ┌─────────────────────────────────────────────────────────────────────────────┐\n │ 完成 │\n └─────────────────────────────────────────────────────────────────────────────┘\n```\n\n### 复杂度模型\n\n| 任务 | 复杂度 | 代理数量 | 验证者 |\n| ---------------------- | ---------- | -------- | ------------------------------------------------- |\n| 修复 README 中的错别字 | 简单 | 1 | 无 |\n| 添加暗黑模式切换按钮 | 简单 | 2 | 通用验证者 |\n| 重构认证系统 | 标准 | 4 | 需求验证、代码验证 |\n| 实现支付流程 | 关键 | 7 | 需求验证、代码验证、安全性验证、测试员验证、对抗测试 |\n\n### 按复杂度选择模型\n\n| 复杂度 | 计划者 | 工作者 | 验证者 |\n| ---------- | -------- | ------- | ------------ |\n| 极简 | - | level1 | - |\n| 简单 | - | level2 | 1 (level2) |\n| 标准 | level2 | level2 | 2 (level2) |\n| 关键 | level3 | level2 | 5 (level2) |\n\n等级对应于特定提供商的模型。可通过 `zeroshot providers setup \u003Cprovider>` 或 `settings.providerSettings` 进行配置。（旧版 `maxModel` 仅适用于 Claude。）\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>自定义工作流（框架模式）\u003C\u002Fstrong>\u003C\u002Fsummary>\n\nZeroshot 是基于消息驱动的，因此您可以定义任意代理拓扑结构。\n\n- 专家小组：并行的专家 -> 聚合器 -> 决策\n- 分阶段审核：顺序验证者，每个都有否决权\n- 层次化：主管动态生成工作者\n- 动态调整：指挥官在执行过程中添加代理\n\n**协调原语：**\n\n- 消息总线（发布\u002F订阅主题）\n- 触发器（条件触发唤醒代理）\n- 账本（SQLite，崩溃恢复）\n- 动态生成（CLUSTER_OPERATIONS）\n\n#### 使用提供商 CLI 创建自定义集群\n\n启动您的提供商 CLI 并描述您的集群：\n\n```\n为安全关键功能创建一个 zeroshot 集群配置：\n\n1. 实现代理（level2）负责实现该功能\n2. 四个并行验证者：\n - 安全验证者：OWASP 检查、SQL 注入、XSS、CSRF\n - 性能验证者：避免 N+1 查询、正确索引\n - 隐私验证者：GDPR 合规、数据最小化\n - 代码审查员：整体代码质量\n3. 所有验证者必须批准后才能合并\n4. 如果任何验证者拒绝，实现代理需修复并重新提交\n5. 安全验证者使用 level3（风险最高）\n\n请参考 cluster-templates\u002Fbase-templates\u002Ffull-workflow.json，并创建类似的集群，保存为 cluster-templates\u002Fsecurity-review.json\n```\n\n内置验证会检查缺失的触发器、死锁以及无效的类型连接，在运行前确保无误。\n\n有关集群模式和示例，请参阅 [CLAUDE.md](.\u002FCLAUDE.md)。\n\n\u003C\u002Fdetails>\n\n## 崩溃恢复\n\n所有状态都持久化在 SQLite 账本中。您可以在任何时候恢复：\n\n```bash\nzeroshot resume cluster-bold-panther\n```\n\n## 隔离模式\n\n### Git 工作树（默认用于 --pr\u002F--ship）\n\n```bash\nzeroshot run 123 --worktree\n```\n\n使用 git 工作树进行轻量级隔离。它会创建一个独立的工作目录和分支。当使用 `--pr` 和 `--ship` 时会自动启用。\n\n### Docker 容器\n\n```bash\nzeroshot run 123 --docker\n```\n\n在一个全新的容器中实现完全隔离。您的工作区将保持不变。适合进行高风险实验或并行运行。\n\n### 何时使用哪种模式\n\n| 场景 | 推荐方式 |\n| ------------------------------------ | ---------------------- |\n| 快速任务，自行审查更改 | 无隔离（默认） |\n| PR 流程，代码审查 | `--worktree` 或 `--pr` |\n| 高风险实验，可能破坏环境 | `--docker` |\n| 并行运行多个任务 | `--docker` |\n| 完全自动，无需审查 | `--ship` |\n\n**默认行为：** 代理仅修改文件，不会提交或推送，除非使用明确允许此操作的隔离模式。\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>Docker 凭证挂载\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n使用 `--docker` 时，zeroshot 会挂载凭证目录，以便代理可以访问提供商 CLI 和 AWS、Azure、kubectl 等工具。\n\n**默认挂载**: `gh`, `git`, `ssh`（GitHub CLI、git 配置、SSH 密钥）\n\n**可用预设**: `gh`, `git`, `ssh`, `aws`, `azure`, `kube`, `terraform`, `gcloud`, `claude`, `codex`, `gemini`\n\n```bash\n# 通过设置配置（持久化）\nzeroshot settings set dockerMounts '[\"gh\", \"git\", \"ssh\", \"aws\", \"azure\"]'\n\n# 查看当前配置\nzeroshot settings get dockerMounts\n\n# 每次运行覆盖\nzeroshot run 123 --docker --mount ~\u002F.aws:\u002Froot\u002F.aws:ro\n\n# 提供商凭证\nzeroshot run 123 --docker --mount ~\u002F.config\u002Fcodex:\u002Fhome\u002Fnode\u002F.config\u002Fcodex:ro\nzeroshot run 123 --docker --mount ~\u002F.config\u002Fgemini:\u002Fhome\u002Fnode\u002F.config\u002Fgemini:ro\n\n# 禁用所有挂载\nzeroshot run 123 --docker --no-mounts\n\n# CI：环境变量覆盖\nZEROSHOT_DOCKER_MOUNTS='[\"aws\",\"azure\"]' zeroshot run 123 --docker\n```\n\n有关提供商 CLI 设置和挂载详情，请参阅 `docs\u002Fproviders.md`。\n\n**自定义挂载**（混合预设与显式路径）：\n\n```bash\nzeroshot settings set dockerMounts '[\n \"gh\",\n \"git\",\n {\"host\": \"~\u002F.myconfig\", \"container\": \"$HOME\u002F.myconfig\", \"readonly\": true}\n]'\n```\n\n**容器主目录**：预设使用 `$HOME` 占位符。默认：`\u002Froot`。可通过以下方式覆盖：\n\n```bash\nzeroshot settings set dockerContainerHome '\u002Fhome\u002Fnode'\n# 或每次运行：\nzeroshot run 123 --docker --container-home \u002Fhome\u002Fnode\n```\n\n**环境变量传递**：预设会自动传递相关环境变量（例如，`aws` -> `AWS_REGION`、`AWS_PROFILE`）。可添加自定义：\n\n```bash\nzeroshot settings set dockerEnvPassthrough '[\"MY_API_KEY\", \"TF_VAR_*\"]'\n```\n\n\u003C\u002Fdetails>\n\n## 资源\n\n- [CLAUDE.md](.\u002FCLAUDE.md) - 架构、集群配置模式、代理原语\n- `docs\u002Fproviders.md` - 提供商设置、模型等级和 Docker 挂载\n- `docs\u002Fcontext-management.md` - 上下文选择、上下文包和状态快照\n- [Discord](https:\u002F\u002Fdiscord.gg\u002FPdZ3UEXB) - 支持与社区\n- `zeroshot export \u003Cid>` - 将对话导出为 Markdown\n- `sqlite3 ~\u002F.zeroshot\u002F*.db` - 直接访问账本以进行调试\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>故障排除\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n| 问题 | 解决方法 |\n| ----------------------------- | ----------------------------------------------------------------------------------------- |\n| `claude: command not found` | `npm i -g @anthropic-ai\u002Fclaude-code && claude auth login` |\n| `codex: command not found` | `npm i -g @openai\u002Fcodex && codex login` |\n| `gemini: command not found` | `npm i -g @google\u002Fgemini-cli && gemini auth login` |\n| `gh: command not found` | [安装 GitHub CLI](https:\u002F\u002Fcli.github.com\u002F) |\n| `--docker` 失败 | Docker 必须正在运行：使用 `docker ps` 验证 |\n| 集群卡住 | 使用 `zeroshot resume \u003Cid>` 继续 |\n| 代理持续失败 | 检查 `zeroshot logs \u003Cid>` 以查看实际错误 |\n| `zeroshot: command not found` | `npm install -g @covibes\u002Fzeroshot` |\n| 代理行为异常 | 在 Claude Code 中使用 `\u002Fanalyze-cluster-postmortem \u003Cid>`（如果修复方案具有通用性则会创建问题） |\n\n\u003C\u002Fdetails>\n\n## 贡献\n\n请参阅 [CONTRIBUTING.md](CONTRIBUTING.md)，了解开发环境搭建及贡献指南。\n\n请在参与之前阅读 [CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md)。\n\n如遇安全相关问题，请参阅 [SECURITY.md](SECURITY.md)。\n\n## TUI\n\nRatatui（Rust 实现）是目前唯一支持的 TUI。入口点如下：\n\n- `zeroshot`（TTY 环境，无需参数）\n- `zeroshot tui`\n- `zeroshot watch`\n\n### TUI 开发\n\nRust 版本的 TUI 会通过标准输入输出启动一个 Node 后端。迭代时请同时运行这两部分。\n\n单命令开发循环（自动构建并重启）：\n\n```bash\ncargo install cargo-watch\nnpm run dev:tui\n```\n\n1. 安装依赖\n\n ```bash\n npm ci\n ```\n\n2. 以监听模式构建 TUI 后端\n\n ```bash\n npm run build:tui-backend -- --watch\n # 或者\n npx tsc -p tsconfig.tui-backend.json -w\n ```\n\n3. 运行 Rust 版本的 TUI（在另一个终端中）\n\n ```bash\n cd tui-rs\n cargo run -p zeroshot-tui -- --ui disruptive\n ```\n\n### 从本仓库本地运行 CLI\n\n若希望全局使用开发分支中的 `zeroshot` 命令：\n\n```bash\nnpm run dev:link\n```\n\n一条命令即可完成链接并启动 TUI 开发循环：\n\n```bash\nnpm run dev:bootstrap\n```\n\n### CLI 集成流程\n\n使用 Node CLI 启动本地的 Rust 二进制文件：\n\n```bash\ncd tui-rs\ncargo build -p zeroshot-tui\ncd ..\nZEROSHOT_TUI_BINARY_PATH=\"$PWD\u002Ftui-rs\u002Ftarget\u002Fdebug\u002Fzeroshot-tui\" node cli\u002Findex.js tui\n```\n\n### TUI 自定义配置\n\n- `ZEROSHOT_TUI_BACKEND_PATH` 可指向特定的 `lib\u002Ftui-backend\u002Fserver.js`\n- `ZEROSHOT_TUI_BINARY_PATH` 可指向本地的 Rust 二进制文件\n- `ZEROSHOT_TUI_UI=classic|disruptive` 可强制指定 UI 风格\n\n---\n\nMIT 许可证 - [Covibes](https:\u002F\u002Fgithub.com\u002Fcovibes)\n\n基于 Anthropic 的 [Claude Code](https:\u002F\u002Fclaude.com\u002Fproduct\u002Fclaude-code) 构建。","# Zeroshot 快速上手指南\n\nZeroshot 是一个开源的 AI 编码代理编排 CLI 工具。它通过运行多智能体工作流（规划者、实现者、独立验证者），在隔离环境中自动执行代码的编写、审查、测试和验证，直到变更被确认或通过可复现的错误报告被拒绝。适用于对代码正确性要求高于速度的场景。\n\n## 环境准备\n\n**系统要求**\n- **操作系统**：Linux 或 macOS。（Windows 用户建议使用 WSL）\n- **Node.js**：版本 18 或更高。\n\n**前置依赖**\n在使用 Zeroshot 之前，你需要安装并登录至少一个 AI 提供商的 CLI 工具，以及对应平台的 Issue 管理工具（如使用 GitHub Issues）。\n\n1. **安装 AI 提供商 CLI**（任选其一或多个）：\n ```bash\n npm i -g @anthropic-ai\u002Fclaude-code\n npm i -g @openai\u002Fcodex\n npm i -g @google\u002Fgemini-cli\n # Opencode 请参考其官方文档\n ```\n\n2. **认证 AI 提供商**：\n ```bash\n claude login # Claude\n codex login # Codex\n gemini auth login # Gemini\n opencode auth login # Opencode\n ```\n\n3. **安装平台 CLI**（用于读取 Issue，以 GitHub 为例）：\n ```bash\n # 安装 GitHub CLI\n gh auth login\n ```\n *注：若使用 GitLab、Jira 或 Azure DevOps，请分别安装 `glab`、`jira` 或 `az` 并完成认证。*\n\n## 安装步骤\n\n使用 npm 全局安装 Zeroshot：\n\n```bash\nnpm install -g @covibes\u002Fzeroshot\n```\n\n> **提示**：如果下载速度较慢，可配置 npm 国内镜像源（如淘宝源）后再执行安装：\n> ```bash\n> npm config set registry https:\u002F\u002Fregistry.npmmirror.com\n> npm install -g @covibes\u002Fzeroshot\n> ```\n\n安装完成后，可通过以下命令查看可用提供商或设置默认提供商：\n```bash\nzeroshot providers\nzeroshot providers set-default codex\n```\n\n## 基本使用\n\nZeroshot 支持直接通过 Issue 编号、Markdown 文件或自然语言描述来启动任务。\n\n**1. 处理 GitHub\u002FGitLab\u002FJira Issue**\n直接输入 Issue 编号（需在对应仓库目录下）或完整 URL：\n```bash\nzeroshot run 123\nzeroshot run https:\u002F\u002Fgithub.com\u002Forg\u002Frepo\u002Fissues\u002F123\nzeroshot run PROJ-789\n```\n\n**2. 基于本地文件运行**\n将任务描述写在 Markdown 文件中运行：\n```bash\nzeroshot run feature.md\n```\n\n**3. 直接输入自然语言指令**\n适合简单任务或快速原型：\n```bash\nzeroshot run \"Add dark mode\"\n```\n\n**4. 复杂任务示例**\n可以直接描述复杂的逻辑需求，Zeroshot 会自动规划并验证：\n```bash\nzeroshot run \"Add optimistic locking with automatic retry: when updating a user,\nretry with exponential backoff up to 3 times, merge non-conflicting field changes,\nand surface conflicts with details. Handle the ABA problem where version goes A->B->A.\"\n```\n\n**常用参数**\n- `--worktree`：在 git worktree 隔离环境中运行。\n- `--docker`：在 Docker 容器中运行（最高隔离级别）。\n- `--pr`：完成任务后自动创建 Pull Request。\n- `--ship`：创建 PR 并在批准后自动合并。\n\n```bash\n# 示例：在 Docker 中运行并自动创建 PR\nzeroshot run 123 --docker --pr\n```","某电商平台的后端团队正面临一个棘手的任务：需要在用户下单流程中紧急引入“乐观锁”机制，以解决高并发下的库存超卖问题，同时必须确保代码能正确处理复杂的版本冲突（如 ABA 问题）。\n\n### 没有 zeroshot 时\n- **人工实现风险高**：资深工程师需手动编写复杂的重试逻辑和版本比对代码，极易因疏忽遗漏边缘情况，导致生产环境出现数据不一致。\n- **验证过程繁琐**：开发者需反复手动运行单元测试和集成测试，难以模拟真实的高并发竞争条件，往往依赖直觉而非数据确认修复效果。\n- **迭代效率低下**：一旦发现测试失败，需重新分析日志、修改代码并再次提交，整个“编码 - 测试 - 修复”循环耗时数小时甚至数天。\n- **缺乏独立审查**：单人开发模式缺乏独立的“第二双眼睛”进行盲测验证，代码质量高度依赖个人状态，难以保证生产级稳定性。\n\n### 使用 zeroshot 后\n- **自主闭环开发**：只需输入“添加带指数退避的乐观锁”指令，zeroshot 自动调度规划器拆解任务，并由实现器在隔离环境中完成编码，精准处理 ABA 问题。\n- **自动化盲测验证**：内置的独立验证器自动运行压力测试和边界检查，不依赖开发者主观判断，直到所有测试通过才停止迭代，确保逻辑严密。\n- **无限循环直至达标**：若首次提交未通过验证，zeroshot 会自动分析失败原因并重新修改代码，无需人工干预，直到生成可合并的生产级代码。\n- **可复现的交付成果**：整个过程的状态持久化保存，即使意外中断也能恢复，最终交付的代码附带明确的验证报告，让团队放心合并。\n\nzeroshot 将原本需要数天的人工试错过程，转化为一个自动运行、自我修正且结果可验证的闭环工程流，让正确性不再依赖运气。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fcovibes_zeroshot_bb271f40.gif","covibes","Covibes","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fcovibes_ffb75927.png","The multiplayer AI workspace.",null,"https:\u002F\u002Fcovibes.ai","https:\u002F\u002Fgithub.com\u002Fcovibes",[81,85,89,93,97,101],{"name":82,"color":83,"percentage":84},"JavaScript","#f1e05a",78.2,{"name":86,"color":87,"percentage":88},"Rust","#dea584",16.7,{"name":90,"color":91,"percentage":92},"TypeScript","#3178c6",3.4,{"name":94,"color":95,"percentage":96},"Shell","#89e051",1.2,{"name":98,"color":99,"percentage":100},"Python","#3572A5",0.2,{"name":102,"color":103,"percentage":100},"Dockerfile","#384d54",1395,116,"2026-04-06T00:17:38","MIT","Linux, macOS","未说明",{"notes":111,"python":109,"dependencies":112},"该工具是一个基于 Node.js 的 CLI，不支持 Windows 原生环境（需等待后续支持或使用 WSL）。运行前必须安装至少一个 AI 提供商的 CLI（如 Claude Code, Codex, Gemini 等）并完成登录认证。若需对接 GitHub、GitLab、Jira 或 Azure DevOps 的问题追踪系统，还需安装对应的官方命令行工具并认证。内部状态使用 SQLite 持久化。",[113,114,115,116,117,118,119,120],"Node.js 18+","@anthropic-ai\u002Fclaude-code","@openai\u002Fcodex","@google\u002Fgemini-cli","gh (GitHub CLI)","glab (GitLab CLI)","jira CLI","az (Azure CLI)",[35,14,13,122],"插件",[124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141],"ai-agents","cli","developer-tools","llm","agent-orchestration","agentic-workflow","ai-agent","autonomous-agents","claude","codex","coding-assistant","gemini","generative-ai","github-automation","llm-ops","llm-tools","multi-agent","vibecoding","2026-03-27T02:49:30.150509","2026-04-06T21:11:46.894297",[145,150,155,160,164,168],{"id":146,"question_zh":147,"answer_zh":148,"source_url":149},20039,"为什么在 Windows (WSL) 上运行命令时会卡住或无响应？","这通常是由文件系统竞争条件和文件锁定问题引起的。最近的更新已修复了此问题：\n1. 修复了 followClaudeTaskLogs() 中的竞争条件，现在会在检查过期状态前轮询日志文件。\n2. 添加了 cleanStaleLock() 以清除崩溃进程留下的过期锁文件。\n3. 将过期超时时间从 30 秒减少到 5 秒。\n4. 用异步文件锁替换了阻塞 CPU 的自旋等待。\n\n解决方案：请升级到最新版本（包含这些修复），如果问题仍然存在，请重新打开 Issue。","https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F46",{"id":151,"question_zh":152,"answer_zh":153,"source_url":154},20040,"如何在 `zeroshot attach` 中查看实时的 CPU、内存和网络监控指标？","该功能已在最新版本中完全实现。运行 `zeroshot attach` 时，底部状态栏会实时显示每个 Worker 的指标：\n- CPU 使用率百分比\n- 内存占用 (MB)\n- 网络 I\u002FO (每秒读写字节数)\n\n指标每 1 秒更新一次。如果在非 TTY 环境（如管道输出、CI\u002FCD 环境）或守护进程模式 (-d) 下运行，监控栏会自动禁用以避免干扰日志。macOS 用户也能获得完整支持（通过 ps 和 lsof 命令获取数据）。","https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F15",{"id":156,"question_zh":157,"answer_zh":158,"source_url":159},20041,"如何优化验证器触发机制以避免对未完成工作的计算浪费？","系统已引入“完成质量信号”机制来解决此问题：\n1. Worker 现在会在消息中包含 `completionStatus` 字段（含 `canValidate` 布尔值）进行自我评估。\n2. 验证器增加了逻辑判断：仅当 `message.content?.data?.completionStatus?.canValidate === true` 时才触发。\n3. 如果 `canValidate` 为 false，Worker 会自动重新触发自身继续工作，而不是立即调用昂贵的验证器。\n\n此方案无需大幅修改架构，利用现有的逻辑脚本即可生效，已在 v5.4.0 版本中发布。","https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F60",{"id":161,"question_zh":162,"answer_zh":163,"source_url":154},20042,"在非 TTY 环境或 CI\u002FCD 流程中使用 zeroshot 会有什么表现？","系统会自动检测环境并禁用依赖终端特性的功能（如实时状态栏），以防止 ANSI 转义码污染日志文件。具体逻辑如下：\n- 如果 `process.stdout.isTTY` 为 false\n- 或者检测到 `process.env.CI` 环境变量\n- 或者 `$TERM` 设置为 'dumb'\n\n在上述情况下，状态 footer 将不会渲染，确保日志纯净且兼容 GitHub Actions 等 CI 环境。",{"id":165,"question_zh":166,"answer_zh":167,"source_url":154},20043,"macOS 上如何支持进程指标监控？","macOS 上通过不同的 API 实现完整支持：\n- CPU 和内存：使用 `ps -p \u003Cpid> -o %cpu,rss` 命令获取。\n- 网络连接：使用 `lsof -i -n -P -a -p \u003Cpid>` 命令解析活跃连接数。\n\n代码会自动识别 `process.platform === 'darwin'` 并调用相应的 Darwin 回退路径，用户无需额外配置即可在 macOS 上看到与 Linux 类似的监控数据。",{"id":169,"question_zh":170,"answer_zh":171,"source_url":154},20044,"终端窗口大小改变时，监控界面会出现重叠或错乱吗？","不会。系统已处理终端调整大小的边缘情况。当检测到 `process.stdout.on('resize')` 事件时，程序会自动重新计算滚动区域并重绘底部状态栏，防止监控信息覆盖正常的 Agent 输出内容。",[173,178,183,188,193,198,203,208,213,218,223,228,233,238,243,248,253,258,263,268],{"id":174,"version":175,"summary_zh":176,"released_at":177},118080,"v5.5.0","## Zeroshot v5.5.0\n\n发布日期：2026年2月19日\n\n本次发布将最新的 `dev` 基线提升至 `main` 分支。\n\n## TUI 状态\nTUI **仍在开发中**。\n- 这 **不是** TUI 的完整版本。\n- TUI 的完整发布说明将在后续的专门版本中提供。\n\n## 本版本中的非 TUI 改进\n\n### 编排与运行时可靠性\n- 为代理添加了引导队列和交付相关逻辑（`src\u002Fagent\u002Fguidance-queue.js`、`src\u002Fmessage-buffer.js`、`src\u002Fguidance-topics.js`）。\n- 添加了状态快照的捕获与发布功能，以更好地监控集群进度（`src\u002Fstate-snapshot.js`、`src\u002Fstate-snapshotter.js`）。\n- 改进了编排器及子集群的行为与恢复路径（`src\u002Forchestrator.js`、`src\u002Fsub-cluster-wrapper.js`、`src\u002Fagent-wrapper.js`）。\n- 改进了已运行会话的附加与输入发送流程（`src\u002Fattach\u002Fsend-input.js`、`src\u002Fattach\u002Findex.js`）。\n\n### 验证工作流加固\n- 添加了分阶段验证模板（`quick-validation`、`heavy-validation`），并升级了 `full-workflow` 模板的行为。\n- 新增了模板模拟与深度验证模块（`src\u002Ftemplate-validation\u002F**`），以便更早地发现触发与共识相关的问题。\n- 加强了配置验证与路由功能（`src\u002Fconfig-validator.js`、`src\u002Fconfig-router.js`、`src\u002Fschemas\u002Fsub-cluster.js`）。\n\n### 提供者与输出的鲁棒性\n- 在各提供者端强化了输出提取以及致命错误\u002F普通错误的解析逻辑（`src\u002Fagent\u002Foutput-extraction.js`）。\n- 添加了提供者重试与退避机制处理（`src\u002Fagent\u002Frate-limit-backoff.js`）。\n- 更新了提供者映射与默认配置，并改进了解析器（`src\u002Fproviders\u002Fopenai\u002F**`、`src\u002Fproviders\u002Fanthropic\u002F**`、`src\u002Fproviders\u002Fgoogle\u002F**`、`src\u002Fproviders\u002Fbase-provider.js`）。\n\n### Git\u002FPR 自动化与工作流行为\n- 增强了 Git 推送器的行为及模式处理（`src\u002Fagents\u002Fgit-pusher-template.js`）。\n- 改进了对 GitHub、GitLab、Azure 和 Jira 等平台问题单提供者的处理能力（`src\u002Fissue-providers\u002F**`）。\n- 优化了工作树隔离处理及 Compose 清理流程（`src\u002Fisolation-manager.js`）。\n\n### CLI 与自动化改进\n- 扩展了 CLI 行为并更新了命令处理逻辑（`cli\u002Findex.js`）。\n- 新增或更新了启动、设置及仓库配置相关的辅助工具（`lib\u002Fstart-cluster.js`、`lib\u002Frepo-settings.js`、`lib\u002Fsettings.js`）。\n- 改进了预检验证流程（`src\u002Fpreflight.js`）。\n\n### CI\u002F发布与质量覆盖\n- 更新了 CI\u002F发布工作流文件（`.github\u002Fworkflows\u002Fci.yml`、`.github\u002Fworkflows\u002Frelease.yml`、`.github\u002Fworkflows\u002Fcodeql.yml`）。\n- 大幅扩展了单元测试与集成测试的覆盖范围，涵盖运行时、验证、提供者及编排器等各个路径（`tests\u002F**`）。\n\n## 发布参考\n- 发布 PR：https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fpull\u002F411\n- 合并提交：`0edf0680b61c346af1960e88a6587e5460ac74fa`","2026-02-19T19:48:38",{"id":179,"version":180,"summary_zh":181,"released_at":182},118081,"v5.4.0","# [5.4.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv5.3.0...v5.4.0) (2026-01-18)\n\n\n### 错误修复\n\n* 添加启动超时，以防止提供者 CLI 卡住时无限挂起 ([57f145f](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F57f145f27fa68ce1cdf3f2f78e7719255d96d18b))\n* **adversarial-tester:** 修复 CANNOT_VALIDATE 滥用漏洞 ([e3bdf31](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fe3bdf313a1ebd898e3559d10e9bb3ebe484ca0ae))\n* **conductor:** 添加负面示例，以防止错误的 CRITICAL 分类 ([#85](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F85)) ([6101fa4](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F6101fa44245e3066d8a77648abbdbd735facca0b))\n* **hooks:** 对替换值中的模板模式进行转义 ([8cf4938](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F8cf4938f16e2d96ea4c65be81731721785f355f7))，关闭了 [#60](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F60)\n* **metrics:** 从任务存储中轮询真实的 PID ([#92](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F92)) ([6afda6d](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F6afda6d4b770fd1245aef799400075fab61518f1))\n* 编排器集群 ID 和 Codex Git 仓库检查 ([#88](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F88)) ([05f224c](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F05f224cd10d5a303e4f2f438a0abce7a59175b04))\n* 编排器集群 ID 和 Codex Git 仓库检查 ([#90](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F90)) ([bace190](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fbace190a78558c72b14e05592d868977145822f3))\n* **output-parser:** 处理 Haiku 的助手消息中的文本块 ([#65](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F65)) ([3b54e67](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F3b54e67b5ff60916ce323a69c2f6f1454ce6d47e))，关闭了 [#52](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F52)\n* **pid:** 在启动后从 tasks.json 中获取真实的 Claude 进程 PID ([c829bb8](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fc829bb8a5ad5b2c41dc4a1b7dfbd70732e88fdf3))\n* 防止提供者 CLI 卡住时无限挂起（opencode 429 漏洞） ([3305da8](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F3305da8df054bfbaeea9145befe74908e965b1f3))\n* **prompts:** 添加泛化检查，使用通用上下文文件 ([#94](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F94)) ([2469f92](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F2469f929710ea0eaa73a15d37fcb1c732ad05b28))\n* **spawn:** 断开 IPC 通道，以便父进程可以退出 ([e89241d](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fe89241d94a257f50b3386fe40fc1dbecf49c2101))\n* **validators:** 使用 testResults 模式强制执行实际执行 ([3431dff](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F3431dff4b9f784ca1702a8150892f7d0150229bf))\n* **validators:** 在重试迭代中跳过 CANNOT_VALIDATE 条件 ([a3c5de3](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fa3c5de33cc4bdbbe2924d3cacc293c29e9706a37))\n\n\n### 新特性\n\n* 添加 Claude API 密钥管理和认证 ([#62](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F62)) ([cf142c8](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fcf142c8a7c82e","2026-01-18T08:13:43",{"id":184,"version":185,"summary_zh":186,"released_at":187},118082,"v5.3.0","# [5.3.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv5.2.1...v5.3.0) (2026-01-12)\n\n\n### Bug修复\n\n* **agent:** 过滤错误提取中的JSON，以防止误匹配 ([b5833aa](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fb5833aac3430f126443c8151cf58ce37a2dd6060))\n* **agents:** 在git-pusher-agent中使用modelLevel代替model ([646da10](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F646da109a25d9ee0a5709d975bdf0bfd52df229d))\n* 清除require-await的lint错误 ([c96c188](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fc96c188d1671813c9e09881619f3e0f66d3f3d39))\n* **conductor:** 增加鲁棒性，防止钩子失败时静默挂起 ([5453958](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F54539588de424fc53438fb51c8f8bd7fbed58768))\n* 在模板加载后注入completion agent ([0d92a36](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F0d92a3658f9301d2f20dc742eb55a37f181689ad))\n* **openai:** Codex CLI支持结构化输出 ([999c55b](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F999c55beb7b80ca5aa87e006af0ffa37259e0231))\n* **orchestrator:** 在前台运行时尊重--pr参数 ([710466a](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F710466aca54e072b464846bda1cba0aef82e4794))\n* 持久化任务状态并恢复过期任务 ([56f51e0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F56f51e01afde171d31b47088493863521c6e351c))\n* **preflight:** 在Windows上使用跨平台命令检测 ([adf599c](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fadf599c240c4a6f6a4be74ad17f6bcb6005a4726)), 关闭了[#47](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F47)\n* **race:** 在检查过期输出之前轮询日志文件，并自动清理过期锁 ([78486e2](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F78486e2627c514cd8d054cb4e158083bb66926a2))\n* 将CPU阻塞式的自旋等待替换为异步文件锁 ([f5bfb0d](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Ff5bfb0d80a2476d6e38a118c4c5dc4202d9a5921))\n* **task-executor:** 处理监视器崩溃时的过期任务状态 ([026357b](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F026357b5e2af9b3cb19349c28825bb5a38b207da))\n* **testers:** 要求行为测试，而非代码审查 ([27f9e08](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F27f9e08d21055b2a6d58a5121bbe69fe12909893)), 关闭了[#828](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F828)、[#829](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F829)和[#830](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F830)\n* **validator:** 将零长度的回退识别为有效的角色处理 ([474ca9b](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F474ca9b217518a23d7895c0872108b8a6599349c))\n\n\n### 新特性\n\n* 添加minModel设置和Markdown文件输入 ([e689472](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fe6894721776c76c7023c7a87d63d132c9e694568)), 关闭了[#42](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F42)\n* 添加模型覆盖功能，并修复--pr完成代理 ([36fb714](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F36fb714fd24265cca2054a9727c32555d4a99a0f))\n* 添加多提供商CLI支持（claude\u002Fcodex\u002Fgemini） ([301fc5f](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit","2026-01-12T23:24:16",{"id":189,"version":190,"summary_zh":191,"released_at":192},118083,"v5.2.1","## [5.2.1](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv5.2.0...v5.2.1) (2026-01-07)\n\n\n### 错误修复\n\n* **git-pusher:** 处理没有验证节点的集群 ([f4cec4d](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Ff4cec4d179273cb4d46240cad9b3b19be497690e))\n\n\n\n","2026-01-07T22:47:54",{"id":194,"version":195,"summary_zh":196,"released_at":197},118084,"v5.2.0","# [5.2.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv5.1.0...v5.2.0) (2026-01-07)\n\n\n### 错误修复\n\n* **test:** 为 e2e-claude-command 测试添加重试逻辑 ([ed52c86](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fed52c86f86180da02a37616018797acaa58aa764))\n\n\n### 功能特性\n\n* 添加 claudeCommand 配置项，用于自定义 Claude CLI ([#38](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F38)) ([6e1a140](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F6e1a140e75b2c37d2a071b8afd853656ba5e12f7)), 关闭了 [#37](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F37)\n\n\n\n","2026-01-07T18:48:12",{"id":199,"version":200,"summary_zh":201,"released_at":202},118085,"v5.1.0","# [5.1.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv5.0.0...v5.1.0) (2026-01-07)\n\n\n### 功能\n\n* 添加对 root 用户的预检检查 ([802f0f4](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F802f0f41e27c1aa585dec31217fa2aeb3c9ba3db))\n\n\n\n","2026-01-07T08:10:42",{"id":204,"version":205,"summary_zh":206,"released_at":207},118086,"v5.0.0","# [5.0.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv4.2.0...v5.0.0) (2026-01-07)\n\n\n* 功能性更新：移除交互式设置向导 ([5438953](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F54389536e478be1dce0ab707ac7ef8e4ef6ce26d))\n\n\n### 重大变更\n\n* 已移除交互式设置向导。请通过以下命令进行配置：\n zeroshot settings set maxModel opus\n zeroshot settings set dockerMounts '[\"gh\",\"git\",\"ssh\"]'\n\n关闭：设置向导阻碍了非交互式使用场景\n\n\n","2026-01-07T06:53:13",{"id":209,"version":210,"summary_zh":211,"released_at":212},118087,"v4.2.0","# [4.2.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv4.1.4...v4.2.0) (2026-01-06)\n\n\n### 功能特性\n\n* **安全:** 添加自动化安全扫描和依赖项管理 ([7b6ae13](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F7b6ae139e97ddcefa46ab3e264e25f6f609163dd))\n\n\n\n","2026-01-06T13:29:07",{"id":214,"version":215,"summary_zh":216,"released_at":217},118088,"v4.1.4","## [4.1.4](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv4.1.3...v4.1.4) (2026-01-06)\n\n\n### 错误修复\n\n* **ci:** 强制对所有发布版本进行 CI 测试 ([3fad703](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F3fad703fb6eb6b7ac8f98400466bac92329c8561))\n\n\n\n","2026-01-06T13:18:35",{"id":219,"version":220,"summary_zh":221,"released_at":222},118089,"v4.1.3","## [4.1.3](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv4.1.2...v4.1.3) (2026-01-06)\n\n\n### 错误修复\n\n* **package:** 将 `scripts\u002F` 目录包含在 npm 包文件中 ([4bd5991](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F4bd599163df5c7b7f9309f07c4af4b1ec5e7bf38))\n* **preflight:** 支持 macOS Keychain 身份验证检测 ([#35](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F35)) ([a6f0880](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fa6f08807eb2b44b241800a2240a76bf6dbedceca))\n\n\n\n","2026-01-06T12:00:50",{"id":224,"version":225,"summary_zh":226,"released_at":227},118090,"v4.1.2","## [4.1.2](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv4.1.1...v4.1.2) (2026-01-05)\n\n\n### Bug Fixes\n\n* **orchestrator:** prevent subscription race condition (issue [#31](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F31)) ([9a41bf1](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F9a41bf1ba9ccdac0f1653eba9d208c6de4028dbe))\n\n\n\n","2026-01-05T17:12:32",{"id":229,"version":230,"summary_zh":231,"released_at":232},118091,"v4.1.1","## [4.1.1](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv4.1.0...v4.1.1) (2026-01-05)\n\n\n### Bug Fixes\n\n* **docker:** change default containerHome from \u002Froot to \u002Fhome\u002Fnode ([7195539](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F71955397d727126525d36f6964980abd0ac1e94c))\n\n\n\n","2026-01-05T09:03:43",{"id":234,"version":235,"summary_zh":236,"released_at":237},118092,"v4.1.0","# [4.1.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv4.0.0...v4.1.0) (2026-01-05)\n\n\n### Features\n\n* **settings:** improve Docker config display with grouped format ([b743673](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fb743673f32d101bb0b86c1b619d85e856a1378c4)), closes [#19](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F19)\n\n\n\n","2026-01-05T08:53:51",{"id":239,"version":240,"summary_zh":241,"released_at":242},118093,"v4.0.0","# [4.0.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv3.1.0...v4.0.0) (2026-01-04)\n\n\n### Bug Fixes\n\n* adversarial tester condition and README accuracy ([c12109b](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fc12109b5ee574301e472bd09ec7495f3a578dc36))\n* **ci:** use correct agent state in status-footer test ([c6f54a8](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fc6f54a89d91a621a8d92c1a21dfa796743e38cd2))\n* **cli:** ensure PROCESS_SPAWNED sets EXECUTING_TASK state ([4c3cc9c](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F4c3cc9c82b67513cf6ab5e5eca9de1b6d259a9d1))\n* **ledger:** prevent write-after-close race condition ([6b64fcf](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F6b64fcfa37a4396591599774c788a022cdbfb1e9))\n* **release:** allow semantic-release to query remote tags ([0be475b](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F0be475b264d400c6b504306e7c535b2736dfaaa1))\n* **release:** explicitly fetch tags for semantic-release ([cecf735](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fcecf7358d9091992d4c7a1191f874588ba7a592d))\n* **tests:** ensure first-run tests are isolated from module cache ([e55dbe7](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fe55dbe7255bab7cf3ec4ddefcc897ec71296a74a))\n* **tests:** move env var and module setup to before() hook ([cf787ff](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fcf787ff7453d1a65cbaaf98655606ccb38dea967))\n* **tests:** use validateConfig for modelRules catch-all validation ([4092d78](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F4092d78be5739f6a3ca4bc80b3dc25ea7c41f74d))\n\n\n### chore\n\n* bump version to 4.0.0 ([95844e8](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F95844e8ffeee4d24dde56b084053d0cdcd30d3e9))\n\n\n### Features\n\n* **context:** enforce maximum informativeness, minimum verbosity ([f99a7b7](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Ff99a7b738214863744119a9b96a50590034299aa))\n* **prompts:** add universal language\u002Ftask support with LLM antipattern detection ([906102b](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F906102b654914ccd73ebb8abaa121304ee4f347e))\n\n\n### Performance Improvements\n\n* **ci:** reduce matrix from 6 jobs to 1 (save ~90% minutes) ([cad652d](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fcad652d22fdc24cf10efabf04e13902529c05b98))\n* **validators:** remove relevance\u002Fnotes fields to save tokens ([b775e5a](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fb775e5a028475f2f11d3b87ec0202c4398100c1d))\n\n\n### BREAKING CHANGES\n\n* CREW_* env vars renamed to ZEROSHOT_*\n\n🤖 Generated with [Claude Code](https:\u002F\u002Fclaude.com\u002Fclaude-code)\n\nCo-Authored-By: Claude Opus 4.5 \u003Cnoreply@anthropic.com>\n* **prompts:** Validator prompts no longer include language-specific examples\n\n\n\n","2026-01-04T11:18:21",{"id":244,"version":245,"summary_zh":246,"released_at":247},118094,"v3.1.0","# [3.1.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv3.0.0...v3.1.0) (2026-01-03)\n\n\n### Bug Fixes\n\n* **attach:** detect cluster IDs without prefix by checking clusters.json ([a3f3b3a](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fa3f3b3a1c3de47333297b98327f36aefb36cb958))\n* **cli:** use canonical AGENT_STATE constants for status footer ([ac53f83](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fac53f83b0af9a7f2de8264ca791457e4e0afca9a))\n* **footer:** show agents during evaluating_logic, building_context, executing_task ([f3c3484](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Ff3c348400d4b2e960410121cef0614dc583e7528))\n* handle Claude CLI lock contention in parallel validators ([b88d502](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fb88d502699c8c0e628310a9b998c4a1f4cb26d1a))\n* **orchestrator:** add missing close() method for test cleanup ([a642886](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fa6428867647de4d5b28c61163be87d711e001c7c))\n* **output:** broadcast text output, not just JSON ([adc8556](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fadc8556a47e3d02ed7189d8290e9cf81a07c909c))\n* **output:** change from MINIMAL to INFORMATIVE output style ([3b87466](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F3b87466eb0012f089edac2fb66a4d118a39e92e0))\n* **planner:** explicitly forbid Deferred and Why defer patterns ([0504b0a](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F0504b0a5e4e4cd0c902989e29a3759fbc46aa534))\n* **planner:** forbid scope reduction in planner prompt ([a9dbfb2](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fa9dbfb2bef67a14289af91b0cea77880fe3eff3f))\n* **planner:** prevent silent phase omission in scope reduction checks ([7e99787](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F7e99787593cd970b45901f7cf1bf641bf4e5f772))\n* **status-footer:** cleanup footer on stop regardless of hidden state ([52fe9e9](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F52fe9e9efeae5c768291a9a0810399bfdae03934))\n* **templates:** hardcode completion-detector model to haiku ([78b917e](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F78b917e1c97697bf64d7a0897c2b68d8bb0bbaa3))\n* **tests:** set ZEROSHOT_WORKTREE env in git-safety-hook tests ([7399cfc](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F7399cfca1d42f748314db355eb247e426fad97a2))\n* **tests:** skip isolation tests when Docker image unavailable ([142f43c](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F142f43c6af6e209a95b286556d13bb594985e850))\n* **tests:** update settings test for maxModel rename and fix git hook case sensitivity ([6cbb654](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F6cbb654fd2ef15fde9f1454d63cd6aae6807404b))\n* **tests:** update tests for maxModel cost ceiling rename ([45b4ac8](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F45b4ac809c480205345be96249608ea2b284f50e))\n* **update-checker:** check npm write permissions before auto-update ([dd9efa8](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fdd9efa83edeef812f6d0ad6142a8e8c7ec4006e6))\n* **watcher:** add global error handlers to prevent silent crashes ([cea4b57](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fcea4b57fe7cfea899bf8981c2b0d200d1c0a9050))\n* **worker:** forbid scope reduction excuses in worker prompt ([c666847](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fc6668473c7f2882482b0593950db780088721925))\n* **worktree:** inject cwd into dynamically spawned template agents ([4c3b916](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F4c3b9162e5656133b01ccbf58c91782855669e33))\n\n\n### Features\n\n* **agents:** conditional git restriction based on isolation mode ([70eb368](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F70eb3681c3d55747d72b491a4e85279b0e215ab5))\n* **orchestrator:** persist agent runtime states for accurate status display ([4205c7d](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F4205c7d0234d3e34e0000ed15ac218c9edb7d048))\n* **validation:** enforce E2E verification with technical constraints ([f2a680a](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Ff2a680ada66e1485d174084d346c0ae9932ce2c9))\n* **worker:** add aggressive COMPLETION MINDSET to worker prompts ([0c6e37b](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F0c6e37b4c0c58cab8b77b7ed1ba23ebb73f55d29))\n\n\n\n","2026-01-03T18:26:52",{"id":249,"version":250,"summary_zh":251,"released_at":252},118095,"v3.0.0","# [3.0.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv2.1.0...v3.0.0) (2025-12-29)\n\n\n### Bug Fixes\n\n* **isolation:** replace busy-wait with async\u002Fawait for parallel copy ([c8afbf0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fc8afbf00927ce939af633406c47a928507c339c4)), closes [#21](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F21)\n* **security:** escape shell arguments in Docker commands ([43476ad](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F43476adfb3c67634d478b4dd53d52a6afb42b297))\n* shell injection prevention and test reliability improvements ([45254f7](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F45254f7f75b027ba43f6e16fa3668960d4b77f97))\n* **status-footer:** use decimal display for interpolated metrics ([#26](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F26)) ([73ce673](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F73ce67376078f97faefe6724e32ff34619f33374))\n\n\n### Features\n\n* **cli:** change default model ceiling to opus ([#28](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F28)) ([1810be3](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F1810be3a6a2cbfbb4d3aefa711c32f9ff9718f5a))\n* **cli:** change default model ceiling to opus + fix worktree flag cascade ([#29](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F29)) ([eaa30b0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Feaa30b06baf381c4fb7306d08fcd2d4e980de002))\n* **cli:** consolidate StatusFooter for logs -f mode + add blinking agent indicator ([fe2722d](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Ffe2722d157e04048b56368e2c0ffcd7052604f36))\n* real-time metrics via interpolation + maxModel cost ceiling ([#24](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F24)) ([f1db466](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Ff1db46691eca592de67e399aca18f6db3e94d628)), closes [#21](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F21)\n* **settings:** replace defaultModel with maxModel cost ceiling ([#25](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F25)) ([9877dad](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F9877dadad890f78b3af1404b0341da392f6f4bb7)), closes [#23](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F23)\n* **validation:** add Phase 5 template variable validation ([#27](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F27)) ([5e5e7c6](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F5e5e7c6ab2a11ba23a3600d101a9c9c7de02569e))\n\n\n### Performance Improvements\n\n* **isolation:** optimize startup with 4 key improvements ([f28f89c](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Ff28f89c36ac98c341484124bbaffee745818dffa)), closes [#20](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F20) [#21](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F21) [#22](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F22) [#23](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F23) [#20](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F20) [#21](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F21) [#22](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F22) [#23](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F23)\n\n\n### BREAKING CHANGES\n\n* None\n* **settings:** defaultModel setting renamed to maxModel\n* defaultModel setting renamed to maxModel\n\n* feat(status-footer): implement real-time metrics via interpolation\n\nReplace blocking 1s metrics polling with background sampling + interpolation:\n- Sample actual metrics every 500ms (non-blocking background)\n- Display updates every 100ms (10 fps - appears continuous)\n- Values smoothly drift toward targets via lerp (15% per tick)\n- CPU and RAM interpolate; Network is cumulative (no interpolation)\n\nResult: Real-time seeming monitoring while reducing actual polling.\n\n🤖 Generated with [Claude Code](https:\u002F\u002Fclaude.com\u002Fclaude-code)\n\nCo-Authored-By: Claude Opus 4.5 \u003Cnoreply@anthropic.com>\n\n* feat(debug-workflow): harden investigator\u002Ffixer\u002Ftester for senior-dev quality\n\nImplements 7 hardening changes to ensure debug-workflow produces\ntrustworthy output without manual code review:\n\n**Investigator:**\n- Structured rootCauses schema requiring proof each is fundamental\n- Mandatory similarPatternLocations field from codebase-wide scan\n- Prompt requires documenting WHY each cause is root (not symptom)\n\n**Fixer:**\n- Mandatory root cause mapping (each cause → specific fix)\n- Mandatory test addition with escape hatch for valid justifications\n- Must fix ALL similar pattern locations, not just original failure\n\n**Tester:**\n- Structured verification schema with commandResult, rootCauseVerification,\n similarLocationVerification, testVerification, regressionCheck\n- Comprehensive checklist: A (command), B (root causes), C (similar locs),\n D (test quality), E (regression via smart tiering)\n- Explicit forbidden rationalizations and approval criteria\n\nResult: Workflow now blocks incomplete work, band-aid fixes, missing tests,\nand ignored similar bugs. Output can be trusted.\n\n🤖 Generated with [Claude Code](https:\u002F\u002Fclaude.com\u002Fclaude-code)\n\nCo-Authored-By: Claude Opus 4.5 \u003Cnoreply@anthropic.com>\n\n\n\n","2025-12-29T19:33:34",{"id":254,"version":255,"summary_zh":256,"released_at":257},118096,"v2.1.0","# [2.1.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv2.0.0...v2.1.0) (2025-12-29)\n\n\n### Features\n\n* add first-run wizard and update checker ([c93cdfe](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fc93cdfec65f05ce5b1ed4583b5ec0a23fcf56f31)), closes [#17](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F17)\n\n\n\n","2025-12-29T10:50:35",{"id":259,"version":260,"summary_zh":261,"released_at":262},118097,"v2.0.0","# [2.0.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv1.5.0...v2.0.0) (2025-12-29)\n\n\n### Code Refactoring\n\n* **cli:** simplify flag hierarchy with cascading --ship → --pr → --isolation ([#18](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F18)) ([5718ead](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F5718ead37f1771a5dfa68dd9b4f55f73e1f6b9d7)), closes [#17](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F17)\n\n\n### BREAKING CHANGES\n\n* **cli:** `auto` command removed, use `run --ship` instead\n\n- Remove `auto` command (use `run --ship` for full automation)\n- Add `--ship` flag: isolation + PR + auto-merge\n- `--pr` now auto-enables `--isolation`\n- Rename `clear` → `purge` for clarity\n- Update help text with cascading flag examples\n- Add `agents` command stubs\n- Add `--json` output support to list\u002Fstatus\n\n\n\n","2025-12-29T10:45:23",{"id":264,"version":265,"summary_zh":266,"released_at":267},118098,"v1.5.0","# [1.5.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv1.4.0...v1.5.0) (2025-12-28)\n\n\n### Bug Fixes\n\n* **agent:** stop polling after max failures and fix status matching ([7a0fbfe](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F7a0fbfe5439f428bdf8e0bcadbd308542221b6f1))\n* **ci:** skip pre-push hook in CI environment ([352b013](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F352b013b71fcea7c2d484c8274fe7c42139c65ea))\n* **cli:** prevent terminal garbling with status footer coordination ([2716ce5](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F2716ce55eae9a08107788200ab798c3f76815820))\n* **config:** add timeout: 0 default to agent configuration ([6ff66c0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F6ff66c093bf8dfd5048b468fbd250cbfc0d9dbc1))\n* **deps:** regenerate package-lock.json for jscpd dependencies ([c46d84c](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fc46d84c3fc1fbb3d00585922613ff36d829d917a))\n* **infra:** improve container cleanup and npm install robustness ([6c04b46](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F6c04b46374bd3041a8b5c185ca163009f2fb6635))\n* **orchestrator:** prevent 0-message clusters on SIGINT during init ([33ed8f9](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F33ed8f9b90d92da6bf9caf2a7b5e52eadbcecc9f))\n* **template-resolver:** apply param defaults before resolving placeholders ([eafdd62](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Feafdd62fce381a7b9a7cb9787cd06e13f421171b))\n* **templates:** add timeout parameter to all base templates ([f853ed3](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Ff853ed39e0e566afaf31040ce94923a2dcc7bfb9))\n\n\n### Features\n\n* **agents:** add git prohibition + minimal output instructions ([6f6496c](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F6f6496c5db29073ebbeb6229ac128a5f62d7591f))\n* mechanical enforcement of 7 antipatterns ([4286091](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F428609163f9405a8d4b9e84adaee0edbc6bbb7d1)), closes [#1](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F1) [#5](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F5) [#2](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F2) [#4](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F4) [#3](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F3) [#7](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fissues\u002F7) [covibes\u002Fcovibes#635](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fcovibes\u002Fissues\u002F635)\n* **templates:** add timeout parameter to worker-validator agents ([ee8b17b](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002Fee8b17bc76aa29bb692965fddbc5a993749f11f9))\n\n\n\n","2025-12-28T23:25:26",{"id":269,"version":270,"summary_zh":271,"released_at":272},118099,"v1.4.0","# [1.4.0](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcompare\u002Fv1.3.0...v1.4.0) (2025-12-28)\n\n\n### Features\n\n* **status-footer:** atomic writes + token cost display ([7baf0c2](https:\u002F\u002Fgithub.com\u002Fcovibes\u002Fzeroshot\u002Fcommit\u002F7baf0c228dd5f3489013f75a1782abe6cbe39661))\n\n\n\n","2025-12-28T18:40:13"]