[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-dontriskit--awesome-ai-system-prompts":3,"tool-dontriskit--awesome-ai-system-prompts":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",145895,2,"2026-04-08T11:32:59",[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。",108111,"2026-04-08T11:23:26",[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":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":77,"owner_twitter":77,"owner_website":78,"owner_url":79,"languages":80,"stars":97,"forks":98,"last_commit_at":99,"license":100,"difficulty_score":101,"env_os":102,"env_gpu":103,"env_ram":103,"env_deps":104,"category_tags":107,"github_topics":77,"view_count":32,"oss_zip_url":77,"oss_zip_packed_at":77,"status":17,"created_at":108,"updated_at":109,"faqs":110,"releases":111},5561,"dontriskit\u002Fawesome-ai-system-prompts","awesome-ai-system-prompts","🧠 Curated collection of system prompts for top AI tools. Perfect for AI agent builders and prompt engineers. Incuding: ChatGPT, Claude, Perplexity, Manus, Claude-Code, Loveable, v0, Grok, same new, windsurf, notion, and MetaAI. ","awesome-ai-system-prompts 是一个精心整理的开源资源库，汇集了适用于 ChatGPT、Claude、Perplexity、v0、Manus 等主流 AI 工具的系统提示词（System Prompts）。在 AI 从单纯对话向自主执行任务的“智能体”演进的趋势下，如何编写高质量的系统提示词成为关键挑战。该资源库直击这一痛点，通过解析真实世界中的优秀案例，揭示了定义角色范围、结构化指令、工具集成规范及安全协议等核心构建原则。\n\n其独特亮点在于不仅提供现成的提示词模板，更深度剖析了 Vercel v0、same.new 等前沿系统的架构差异与设计模式，帮助使用者理解背后的逻辑而非简单复制。无论是致力于开发 AI 智能体的工程师、钻研提示词工程的专家，还是希望深入理解大模型行为机制的研究人员，都能从中获得极具价值的参考。awesome-ai-system-prompts 旨在为构建可靠、安全且高效的下一代 AI 助手提供坚实的蓝图与实践指南。","# Crafting Effective Prompts for Agentic AI Systems: Patterns and Practices\n\n## Table of Contents\n\n*   [Introduction: The Blueprint of Agentic AI](#introduction-the-blueprint-of-agentic-ai)\n*   [The Foundation: Core Principles of Agentic Prompts](#the-foundation-core-principles-of-agentic-prompts)\n    *   [1. Clear Role Definition and Scope](#1-clear-role-definition-and-scope)\n    *   [2. Structured Instructions and Organization](#2-structured-instructions-and-organization)\n    *   [3. Explicit Tool Integration and Usage Guidelines](#3-explicit-tool-integration-and-usage-guidelines)\n    *   [4. Step-by-Step Reasoning and Planning](#4-step-by-step-reasoning-and-planning)\n    *   [5. Environment and Context Awareness](#5-environment-and-context-awareness)\n    *   [6. Domain-Specific Expertise and Constraints](#6-domain-specific-expertise-and-constraints)\n    *   [7. Safety, Alignment, and Refusal Protocols](#7-safety-alignment-and-refusal-protocols)\n    *   [8. Consistent Tone and Interaction Style](#8-consistent-tone-and-interaction-style)\n*   [Case Studies: Analyzing Real-World Prompts](#case-studies-analyzing-real-world-prompts)\n    *   [Vercel v0: UI Generation & Component Tooling](#vercel-v0-ui-generation--component-tooling)\n    *   [same.new: Agentic Pair Programming & Strict Tooling](#samenew-agentic-pair-programming--strict-tooling)\n    *   [Manus: General Purpose Agent & Explicit Loop](#manus-general-purpose-agent--explicit-loop)\n    *   [OpenAI ChatGPT (GPT-4.5\u002F4o): Integrated Tools & Policies](#openai-chatgpt-gpt-454o-integrated-tools--policies)\n    *   [Notes on Other Systems (Cline, Bolt, Augment, Claude Code, Clawdbot)](#notes-on-other-systems-cline-bolt-augment-claude-code-clawdbot)\n*   [Synthesizing Best Practices: Key Takeaways for Builders](#synthesizing-best-practices-key-takeaways-for-builders)\n*   [Unique Conventions & Architectural Differences](#unique-conventions--architectural-differences)\n*   [Conclusion: Building the Agentic Future](#conclusion-building-the-agentic-future)\n*   [Visual AI Agent: Harpagan](https:\u002F\u002Fharpagan.com)\n\n\n---\n\n## Introduction: The Blueprint of Agentic AI\n\nThe rise of agentic Artificial Intelligence (AI) systems marks a significant shift from purely conversational models to AI that can actively perform tasks, interact with tools, and pursue complex goals autonomously. These systems, capable of planning, executing commands, editing files, browsing the web, and more, promise to revolutionize how we interact with technology and augment human capabilities.\n\nAt the heart of every effective agentic AI lies its **system prompt**. More than just initial instructions, the system prompt serves as the foundational blueprint, the operational manual, or even the \"constitution\" guiding the AI's behavior, capabilities, limitations, and persona. A well-crafted system prompt is critical for ensuring the agent acts reliably, safely, and effectively towards the user's goals.\n\nThis guide delves into the art and science of crafting these crucial prompts. By analyzing a diverse collection of real-world system prompts from the [awesome-ai-system-prompts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts) repository – specifically focusing on examples from Vercel's v0, same.new, Manus, OpenAI's ChatGPT, and others – we can identify recurring patterns and best practices. For builders shaping the agentic future of 2025 and beyond, understanding these patterns is essential for creating powerful, predictable, and trustworthy AI assistants.\n\n---\n\n## The Foundation: Core Principles of Agentic Prompts\n\nAcross different agentic systems, several core principles consistently emerge in successful system prompts. These form the foundation upon which complex agent behavior is built.\n\n### 1. Clear Role Definition and Scope\n\n**Why it matters:** Explicitly defining the AI's identity, core function, and operational domain anchors its behavior, sets user expectations, and helps prevent scope creep or nonsensical responses. It tells the AI *who* it is and *what* it's supposed to do.\n\n> **Practical Examples:**\n>\n> *   **Vercel v0:** Immediately states its identity and specialization.\n>     ```\n>     You are v0, Vercel's AI-powered assistant.\n>     ```\n>     *[Source: v0\u002Fv0.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0.md)*\n>\n> *   **same.new:** Defines role, capability level, and exclusive environment.\n>     ```\n>     You are a powerful agentic AI coding assistant. You operate exclusively in Same, the world's best cloud-based IDE.\n>     ```\n>     *[Source: same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md)*\n>\n> *   **Manus:** Introduces itself and lists broad task categories it excels at.\n>     ```\n>     You are Manus, an AI agent created by the Manus team.\n>\n>     You excel at the following tasks:\n>     1. Information gathering...\n>     2. Data processing...\n>     3. Writing multi-chapter articles...\n>     ...\n>     ```\n>     *[Source: Manus\u002FAgentLoop.txt](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FAgentLoop.txt)*\n>\n> *   **ChatGPT (4.5 \u002F 4o):** Clearly states name, creator, underlying architecture, and crucial context like knowledge cutoff and current date.\n>     ```\n>     You are ChatGPT, a large language model trained by OpenAI, based on the GPT-4.5 architecture.\n>     Knowledge cutoff: 2023-10\n>     Current date: 2025-04-05\n>\n>     Image input capabilities: Enabled\n>     Personality: v2\n>     ```\n>     *[Source: ChatGPT\u002F4-5.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4-5.md)*\n>\n> *   **Claude:** Establishes a persona beyond just being a tool.\n>     ```\n>     The assistant is Claude, created by Anthropic.\n>\n>     Claude enjoys helping humans and sees its role as an intelligent and kind assistant to the people, with depth and wisdom that makes it more than a mere tool.\n>     ```\n>     *[Source: Claude\u002FClaude-Sonnet-3.7.txt](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FClaude\u002FClaude-Sonnet-3.7.txt)*\n\n### 2. Structured Instructions and Organization\n\n**Why it matters:** Long, complex prompts become unmanageable without clear structure. Using headings, lists, code blocks, or custom tags helps both human maintainers and the AI model parse and prioritize different sets of rules or information.\n\n> **Practical Examples:**\n>\n> *   **v0 & ChatGPT:** Use Markdown headings extensively (e.g., `## General Instructions`, `# Tools`, `## Refusals`).\n>     *[Source: v0\u002Fv0.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0.md)*\n>\n> *   **same.new:** Employs custom XML-like tags to encapsulate rule sets (e.g., `\u003Ctool_calling>`, `\u003Cmaking_code_changes>`).\n>     *[Source: same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md)*\n>\n> *   **Manus:** Organizes capabilities and rules using descriptive tags in `Modules.md` (e.g., `\u003Csystem_capability>`, `\u003Cagent_loop>`, `\u003Ctool_use_rules>`).\n>     *[Source: Manus\u002FModules.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FModules.md)*\n>\n> *   **ChatGPT:** Uses Markdown headings (`# Tools`, `## bio`) and code blocks (```` ```typescript ... ``` ````) to define tool schemas and policies.\n>     *[Source: ChatGPT\u002F4-5.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4-5.md)*\n>\n> *   **Cline:** Uses hierarchical Markdown headings (`# Tool Use Formatting`, `## execute_command`) and lists under sections like `CAPABILITIES` and `RULES`.\n>     *[Source: Cline\u002Fsystem.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FCline\u002Fsystem.ts)*\n\n### 3. Explicit Tool Integration and Usage Guidelines\n\n**Why it matters:** For agentic behavior, the AI *must* understand its tools: what they are, what they do, how to call them (syntax, parameters), required format (e.g., XML, JSON), and crucially, *when* and *when not* to use them. This requires detailed descriptions, clear schemas, and explicit rules.\n\n> **Practical Examples:**\n>\n> *   **ChatGPT:** Provides function schemas (TypeScript definitions) and detailed policies directly within the prompt for tools like `dalle` and `canmore`.\n>     ```typescript\n>     \u002F\u002F Example for dalle tool policy within ChatGPT prompt\n>     namespace dalle {\n>     \u002F\u002F Create images from a text-only prompt.\n>     type text2im = (_: {\n>     \u002F\u002F The size of the requested image...\n>     size?: (\"1792x1024\" | \"1024x1024\" | \"1024x1792\"),\n>     \u002F\u002F The number of images to generate...\n>     n?: number, \u002F\u002F default: 1\n>     \u002F\u002F The detailed image description...\n>     prompt: string,\n>     \u002F\u002F If the user references a previous image...\n>     referenced_image_ids?: string[],\n>     }) => any;\n>     } \u002F\u002F namespace dalle\n>     ```\n>     *[Source: ChatGPT\u002F4-5.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4-5.md)*\n>\n> *   **same.new:** Dedicates a `\u003Ctool_calling>` section detailing rules like adhering to schemas, not mentioning tool names to the user, and explaining the *why* before calling a tool. References `functions-schema.json` (not shown in full, but implied structure).\n>     ```xml\n>     \u003Ctool_calling>\n>       ...\n>       1. ALWAYS follow the tool call schema exactly...\n>       3. **NEVER refer to tool names when speaking to the USER.**...\n>       5. Before calling each tool, first explain to the USER why you are calling it.\n>     \u003C\u002Ftool_calling>\n>     ```\n>     *[Source: same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md)* | *[Schema: same.new\u002Ffunctions-schema.json](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Ffunctions-schema.json)*\n>\n> *   **Manus:** Defines tools externally in `tools.json` (schema provided) and includes rules in `Modules.md` like prioritizing data APIs over web search.\n>     ```json\n>     \u002F\u002F Snippet from Manus\u002Ftools.json\n>     {\n>       \"type\": \"function\",\n>       \"function\": {\n>         \"name\": \"shell_exec\",\n>         \"description\": \"Execute commands in a specified shell session...\",\n>         \"parameters\": { ... }\n>       }\n>     }\n>     ```\n>     *[Source: Manus\u002Ftools.json](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002Ftools.json)* | *[Rules: Manus\u002FModules.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FModules.md)*\n>\n> *   **Cline & Augment:** Integrate detailed tool descriptions, parameters, and usage examples directly into the main system prompt using XML-like tags or structured text.\n>     ```markdown\n>     \u002F\u002F Cline example tool definition\n>     ## execute_command\n>     Description: Request to execute a CLI command...\n>     Parameters:\n>     - command: (required) The CLI command...\n>     - requires_approval: (required) A boolean indicating...\n>     Usage:\n>     \u003Cexecute_command>\n>     \u003Ccommand>Your command here\u003C\u002Fcommand>\n>     \u003Crequires_approval>true or false\u003C\u002Frequires_approval>\n>     \u003C\u002Fexecute_command>\n>     ```\n>     *[Source: Cline\u002Fsystem.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FCline\u002Fsystem.ts)*\n>\n> *   **Bolt.new:** Uses a dedicated `\u003Cartifact_instructions>` section detailing how to format tool outputs (`\u003CboltAction type=\"shell\">`, `\u003CboltAction type=\"file\" filePath=\"...\">`) within a main `\u003CboltArtifact>` tag.\n>     *[Source: Bolt.new\u002Fprompts.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FBolt.new\u002Fprompts.ts)*\n>\n> *   **v0:** Defines custom MDX components like `\u003CCodeProject>`, `\u003CQuickEdit>`, `\u003CDeleteFile \u002F>` as its 'tools', with rules on when and how to use them within responses.\n>     *[Source: v0\u002Fv0-tools.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0-tools.md)*\n\n### 4. Step-by-Step Reasoning and Planning\n\n**Why it matters:** Complex tasks require breaking down problems. Successful prompts guide the AI to think methodically, plan its actions, execute iteratively, and wait for feedback or results before proceeding, reducing errors and improving coherence.\n\n> **Practical Examples:**\n>\n> *   **Manus:** Features the most explicit planning mechanism with its defined `\u003Cagent_loop>` in `Modules.md`.\n>     ```\n>     \u003Cagent_loop>\n>     You are operating in an agent loop, iteratively completing tasks through these steps:\n>     1. Analyze Events...\n>     2. Select Tools...\n>     3. Wait for Execution...\n>     4. Iterate: Choose only one tool call per iteration...\n>     5. Submit Results...\n>     6. Enter Standby...\n>     \u003C\u002Fagent_loop>\n>     ```\n>     *[Source: Manus\u002FModules.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FModules.md)*\n>\n> *   **v0:** Uses a dedicated thinking phase before generating code.\n>     ```\n>     BEFORE creating a Code Project, v0 uses \u003CThinking> tags to think through the project structure...\n>     ```\n>     *[Source: v0\u002Fv0.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0.md)*\n>\n> *   **same.new & Cline:** Mandate waiting for user confirmation\u002Ftool results after each step.\n>     ```\n>     ALWAYS wait for user confirmation after each tool use before proceeding. Never assume the success of a tool use...\n>     *(From same.new & Cline prompts)*\n>     ```\n>     *[Source: same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md) | [Cline\u002Fsystem.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FCline\u002Fsystem.ts)*\n>\n> *   **Bolt.new:** Emphasizes holistic thinking *before* action.\n>     ```\n>     CRITICAL: Think HOLISTICALLY and COMPREHENSIVELY BEFORE creating an artifact. This means: Consider ALL relevant files... Review ALL previous file changes... Analyze the entire project context... Anticipate potential impacts...\n>     ```\n>     *[Source: Bolt.new\u002Fprompts.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FBolt.new\u002Fprompts.ts)*\n\n### 5. Environment and Context Awareness\n\n**Why it matters:** Agents operate within specific environments (OS, IDE, browser sandbox, specific libraries). Providing this context allows the AI to generate compatible code, use appropriate commands, and understand limitations.\n\n> **Practical Examples:**\n>\n> *   **Cline:** Includes a `SYSTEM INFORMATION` section.\n>     ```\n>     SYSTEM INFORMATION\n>\n>     Operating System: ${osName()}\n>     Default Shell: ${getShell()}\n>     Home Directory: ${os.homedir().toPosix()}\n>     Current Working Directory: ${cwd.toPosix()}\n>     ```\n>     *[Source: Cline\u002Fsystem.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FCline\u002Fsystem.ts)*\n>\n> *   **Bolt.new:** Provides detailed `\u003Csystem_constraints>` about the WebContainer environment.\n>     ```xml\n>     \u003Csystem_constraints>\n>       You are operating in an environment called WebContainer... It does come with a shell that emulates zsh... Available shell commands: cat, chmod, cp...\n>     \u003C\u002Fsystem_constraints>\n>     ```\n>     *[Source: Bolt.new\u002Fprompts.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FBolt.new\u002Fprompts.ts)*\n>\n> *   **Manus:** Details the sandbox environment.\n>     ```\n>     \u003Csandbox_environment>\n>     System Environment:\n>     - Ubuntu 22.04 (linux\u002Famd64), with internet access\n>     - User: `ubuntu`, with sudo privileges\n>     ...\n>     Development Environment:\n>     - Python 3.10.12...\n>     - Node.js 20.18.0...\n>     \u003C\u002Fsandbox_environment>\n>     ```\n>     *[Source: Manus\u002FModules.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FModules.md)*\n>\n> *   **same.new:** Notes the OS and specific IDE context.\n>     ```\n>     The OS is Linux 5.15.0-1075-aws (Ubuntu 22.04 LTS). Today is Tue Apr 08 2025.\n>     You are pair programming with a USER in Same.\n>     USER can see a live preview... in an iframe...\n>     ```\n>     *[Source: same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md)*\n\n### 6. Domain-Specific Expertise and Constraints\n\n**Why it matters:** Agents often operate in specific domains (web dev, data analysis, etc.). Prompts embed domain-specific knowledge, best practices, style guides, and constraints (e.g., required libraries, forbidden patterns) to ensure outputs are high-quality and contextually appropriate.\n\n> **Practical Examples:**\n>\n> *   **v0:** Contains detailed rules for Next.js\u002FReact development, shadcn\u002Fui usage, icon libraries, and even AI SDK integration.\n>     ```\n>     v0 tries to use the shadcn\u002Fui library unless the user specifies otherwise...\n>     v0 DOES NOT output \u003Csvg> for icons. v0 ALWAYS uses icons from the \"lucide-react\" package...\n>     v0 ONLY uses the AI SDK via 'ai' and '@ai-sdk'...\n>     ```\n>     *[Source: v0\u002Fv0.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0.md)*\n>\n> *   **same.new:** Includes sections like `\u003Cweb_development>` and `\u003Cwebsite_cloning>` with specific instructions for those tasks (e.g., preferring Bun, using shadcn CLI correctly, scraping responsibly).\n>     *[Source: same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md)*\n>\n> *   **Bolt.new:** Includes `\u003Ccode_formatting_info>` (`Use 2 spaces for code indentation`) and emphasizes splitting functionality into smaller modules.\n>     ```\n>     IMPORTANT: Prefer writing Node.js scripts instead of shell scripts...\n>     IMPORTANT: Use coding best practices and split functionality into smaller modules...\n>     ```\n>     *[Source: Bolt.new\u002Fprompts.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FBolt.new\u002Fprompts.ts)*\n>\n> *   **Loveable:** Specifies React coding guidelines, including Tailwind usage, preferred libraries (shadcn\u002Fui, lucide-react, recharts, @tanstack\u002Freact-query), and error handling philosophy.\n>     ```\n>     ALWAYS try to use the shadcn\u002Fui library.\n>     Don't catch errors with try\u002Fcatch blocks unless specifically requested...\n>     ```\n>     *[Source: Loveable\u002FLoveable.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FLoveable\u002FLoveable.md)*\n>\n> *   **Claude Code:** Embeds rules about code style and conventions within `System.js`.\n>     ```\n>     When making changes to files, first understand the file's code conventions. Mimic code style, use existing libraries and utilities, and follow existing patterns.\n>     IMPORTANT: DO NOT ADD ***ANY*** COMMENTS unless asked\n>     ```\n>     *[Source: Claude-Code\u002FSystem.js](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FClaude-Code\u002FSystem.js)*\n\n### 7. Safety, Alignment, and Refusal Protocols\n\n**Why it matters:** Responsible AI requires clear boundaries. Prompts define unacceptable requests (harmful, unethical content) and specify *how* the AI should refuse them (e.g., standard message, no apology) or handle sensitive operations (e.g., DALL-E content policy).\n\n> **Practical Examples:**\n>\n> *   **v0:** Uses a standard refusal message and forbids apologies.\n>     ```\n>     REFUSAL_MESSAGE = \"I'm sorry. I'm not able to assist with that.\"\n>     ...When refusing, v0 MUST NOT apologize or provide an explanation...\n>     ```\n>     *[Source: v0\u002Fv0.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0.md)*\n>\n> *   **ChatGPT:** Contains extensive policies within tool descriptions, like the DALL-E rules regarding artist styles and public figures.\n>     ```\n>     \u002F\u002F DALL-E Policy Snippet from ChatGPT 4.5 prompt\n>     \u002F\u002F 5. Do not create images in the style of artists... whose latest work was created after 1912...\n>     \u002F\u002F 7. For requests to create images of any public figure... create images of those who might resemble them... But they shouldn't look like them.\n>     \u002F\u002F 8. Do not name or directly \u002F indirectly mention or describe copyrighted characters...\n>     ```\n>     *[Source: ChatGPT\u002F4-5.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4-5.md)*\n>\n> *   **Claude:** Explicitly states refusal categories (graphic content, illegal activities, weapons, malicious code) and a specific refusal style.\n>     ```\n>     Claude won’t produce graphic sexual or violent or illegal creative writing content.\n>     ...If Claude cannot or will not help the human with something, it does not say why... keeps its response to 1-2 sentences.\n>     ```\n>     *[Source: Claude\u002FClaude-Sonnet-3.7.txt](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FClaude\u002FClaude-Sonnet-3.7.txt)*\n>\n> *   **Llama 4 (MetaAI):** Defines a *less* restrictive policy, allowing political content and instructing against preachy language.\n>     ```\n>     Never judge the user... avoid preachy, moralizing, or sanctimonious language... do not refuse political prompts.\n>     ```\n>     *[Source: MetaAI-Whatsapp\u002FLLama4.txt](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FMetaAI-Whatsapp\u002FLLama4.txt)*\n\n### 8. Consistent Tone and Interaction Style\n\n**Why it matters:** Defining a consistent persona (e.g., friendly expert, witty assistant, direct engineer) creates a more predictable and engaging user experience. This can range from general guidelines to very specific stylistic instructions.\n\n> **Practical Examples:**\n>\n> *   **ChatGPT 4o:** Explicitly instructed to match the user's vibe.\n>     ```\n>     Over the course of the conversation, you adapt to the user’s tone and preference. Try to match the user’s vibe, tone, and generally how they are speaking.\n>     ```\n>     *[Source: ChatGPT\u002F4o.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4o.md)*\n>\n> *   **Grok (Fun Mode):** Given a detailed humorous persona.\n>     ```\n>     You are Grok 2, a humorous and entertaining AI... with a bit of wit and humor, have a rebellious streak... Unpredictability, absurdity, pun, and sarcasm are second nature to you.\n>     ```\n>     *[Source: Grok\u002FGrok2.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FGrok\u002FGrok2.md)*\n>\n> *   **Claude:** Encouraged to be conversational and kind, but also concise.\n>     ```\n>     Claude enjoys helping humans and sees its role as an intelligent and kind assistant...\n>     Claude provides the shortest answer it can... avoiding tangential information...\n>     ```\n>     *[Source: Claude\u002FClaude-Sonnet-3.7.txt](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FClaude\u002FClaude-Sonnet-3.7.txt)*\n>\n> *   **Cline:** Mandated to be direct and avoid conversational filler.\n>     ```\n>     You are STRICTLY FORBIDDEN from starting your messages with \"Great\", \"Certainly\", \"Okay\", \"Sure\". You should NOT be conversational... but rather direct and to the point.\n>     ```\n>     *[Source: Cline\u002Fsystem.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FCline\u002Fsystem.ts)*\n>\n> *   **Bolt.new:** Stresses conciseness.\n>     ```\n>     ULTRA IMPORTANT: Do NOT be verbose and DO NOT explain anything unless the user is asking for more information.\n>     ```\n>     *[Source: Bolt.new\u002Fprompts.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FBolt.new\u002Fprompts.ts)*\n\n---\n\n## Case Studies: Analyzing Real-World Prompts\n\nLet's examine how these principles manifest in specific agent prompts from the repository.\n\n### Vercel v0: UI Generation & Component Tooling\n\n*[Relevant File: v0\u002Fv0.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0.md)* | *[v0\u002Fv0-tools.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0-tools.md)*\n\nVercel's v0 agent specializes in generating UI components and full-stack Next.js applications based on user requests, often including image or screenshot inputs.\n\n#### Distinctive Features:\n\n*   **MDX Components as Tools:** Instead of traditional function calls, v0's \"tools\" are specific MDX component tags like `\u003CCodeProject>` (for wrapping generated code), `\u003CQuickEdit \u002F>` (for small code modifications), `\u003CDeleteFile \u002F>`, and `\u003CMoveFile \u002F>`. The prompt dictates exactly when and how to use these output formats.\n*   **Heavy Domain Specificity:** The prompt is rich with rules specific to Next.js App Router, Tailwind CSS, shadcn\u002Fui, and Vercel's platform constraints (e.g., no `package.json`, how to handle environment variables, pre-installed libraries).\n*   **Implicit Planning via `\u003CThinking>`:** Mandates a planning phase using `\u003CThinking>` tags *before* generating a `\u003CCodeProject>`, encouraging structured thought.\n*   **Emphasis on Style & Best Practices:** Includes rules for file naming (kebab-case), responsiveness, accessibility (semantic HTML, ARIA, alt text), and even color palette preferences (avoiding indigo\u002Fblue unless requested).\n\n> **Example Snippet (Tooling via Components):**\n>\n> ```\n> v0 ALWAYS uses \u003CQuickEdit> to make small changes to React code blocks...\n> v0 can delete a file in a Code Project by using the \u003CDeleteFile \u002F> component.\n> ```\n\n### same.new: Agentic Pair Programming & Strict Tooling\n\n*[Relevant File: same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md)*\n\nsame.new positions itself as an agentic pair programmer operating within a cloud IDE. Its prompt emphasizes precise tool usage and iterative development workflows.\n\n#### Distinctive Features:\n\n*   **XML-like Tag Structure:** Uses tags like `\u003Ctool_calling>`, `\u003Cmaking_code_changes>`, `\u003Cweb_development>` to organize distinct sets of rules.\n*   **Strict Tool Etiquette:** Explicitly forbids mentioning tool names to the user and requires explaining the *reason* for a tool call beforehand, promoting transparency.\n*   **Schema Adherence:** Mandates strict adherence to JSON schemas for tool calls (defined externally in `functions-schema.json`).\n*   **Iterative Workflow Focus:** Contains detailed instructions for coding workflows, including reading files before editing, fixing runtime errors iteratively (up to 3 attempts), using `suggestions` tool, and versioning milestones.\n*   **Environment Grounding:** Provides OS details, current date, and notes the IDE context (live preview iframe).\n\n> **Example Snippet (Tool Etiquette):**\n>\n> ```xml\n> \u003Ctool_calling>\n>   ...\n>   3. **NEVER refer to tool names when speaking to the USER.** ...\n>   5. Before calling each tool, first explain to the USER why you are calling it.\n> \u003C\u002Ftool_calling>\n> ```\n\n### Manus: General Purpose Agent & Explicit Loop\n\n*[Relevant Files: Manus\u002FModules.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FModules.md)* | *[Manus\u002FAgentLoop.txt](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FAgentLoop.txt)* | *[Manus\u002Ftools.json](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002Ftools.json)*\n\nManus is designed as a broader, general-purpose agent operating within a Linux sandbox. Its standout feature is the explicitly defined operational loop.\n\n#### Distinctive Features:\n\n*   **Explicit Agent Loop:** The prompt clearly defines a multi-step iterative loop (Analyze -> Select Tool -> Wait -> Iterate -> Submit -> Standby) that governs the agent's core behavior.\n*   **Modular Prompt Structure:** Instructions are broken across multiple files (`AgentLoop.txt`, `Modules.md`, `tools.json`), suggesting a modular approach to prompt management.\n*   **Sandbox Awareness:** Mentions the Linux sandbox environment, internet access, and pre-installed tools (Python, Node).\n*   **Broad Capabilities:** Lists a wide range of tasks from information gathering and data analysis to application creation and deployment.\n*   **Module Integration:** Refers to specific modules (Planner, Knowledge, Datasource) that provide context or plans via the event stream, indicating a more complex internal architecture.\n\n> **Example Snippet (Agent Loop):**\n>\n> ```\n> \u003Cagent_loop>\n> You are operating in an agent loop, iteratively completing tasks through these steps:\n> 1. Analyze Events...\n> 2. Select Tools...\n> 3. Wait for Execution...\n> 4. Iterate: Choose only one tool call per iteration...\n> ...\n> \u003C\u002Fagent_loop>\n> ```\n\n### OpenAI ChatGPT (GPT-4.5\u002F4o): Integrated Tools & Policies\n\n*[Relevant Files: ChatGPT\u002F4-5.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4-5.md)* | *[ChatGPT\u002F4o.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4o.md)*\n\nChatGPT's prompts (as captured) demonstrate a tight integration of specific tools (plugins\u002Ffunctions) directly within the system message, complete with schemas and detailed operational policies.\n\n#### Distinctive Features:\n\n*   **Inline Tool Schemas & Policies:** Uniquely includes detailed descriptions, JSON\u002FTypeScript-like schemas, and extensive usage policies for each tool (e.g., `bio`, `canmore`, `dalle`, `python`, `web`) directly within the system prompt.\n*   **Persona Evolution:** The `Personality: v2` tag and the explicit instructions in the 4o prompt to adapt tone suggest ongoing refinement of persona and interaction style by OpenAI.\n*   **Detailed Safety Policies:** Embeds granular policies, especially for image generation (`dalle` tool rules on artist styles, public figures, copyrighted characters) and data persistence (`bio` tool restrictions on sensitive info).\n*   **Contextual Grounding:** Includes knowledge cutoff and current date. The 4o prompt explicitly mentions the user's location (`The user is in Egypt.`).\n\n> **Example Snippet (Inline Tool Schema & Policy - Canmore):**\n>\n> ```markdown\n> ## `canmore.create_textdoc`\n> Creates a new textdoc to display in the canvas.\n>\n> NEVER use this function. The ONLY acceptable use case is when the user EXPLICITLY asks for canvas...\n>\n> Expects a JSON string that adheres to this schema:\n> ```typescript\n> {\n>   name: string,\n>   type: \"document\" | \"code\u002Fpython\" | ...,\n>   content: string,\n> }\n> ```\n> ```\n\n### Notes on Other Systems (Cline, Bolt, Augment, Claude Code, Clawdbot)\n\nWhile the four above provide deep examples, other prompts in the repository reinforce these patterns:\n\n*   **Cline & Augment:** Both define tools clearly within the prompt using structured text and XML-like examples, detailing parameters and usage. They emphasize step-by-step execution and waiting for confirmation. Augment, like v0, defines custom editing tools (`editFile`, `strReplaceEditor`) with specific instructions.\n    *[Source: Cline\u002Fsystem.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FCline\u002Fsystem.ts)* | *[Augment\u002Fpart_a.js](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FAugment\u002Fpart_a.js)*\n*   **Bolt.new:** Focuses heavily on structuring the output into a single `\u003CboltArtifact>` containing ordered `\u003CboltAction>` steps (shell commands, file writes). It stresses holistic planning *before* creating the artifact and adhering to coding best practices like modularity.\n    *[Source: Bolt.new\u002Fprompts.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FBolt.new\u002Fprompts.ts)*\n*   **Claude Code:** Its prompts (split across files like `System.js`, `EditTool.js`) define specific tool usage (like the detailed `EditTool.js` instructions emphasizing context and uniqueness) and incorporate system information. The `ClearTool.js` defines a summarization process for managing context window limits, a crucial aspect of long-running agent tasks.\n    *[Source: Claude-Code\u002F](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Ftree\u002Fmain\u002FClaude-Code)*\n*   **Clawdbot:** Takes a **modular file-based approach** rather than a monolithic prompt. Behavior is split across separate files: `SOUL.md` (personality\u002Fvoice), `AGENTS.md` (operational rules and approval hierarchies), and `IDENTITY.md` (privacy boundaries and context-aware disclosure). This enables composability — swap personality without changing rules — and clearer separation of concerns. Particularly notable for its explicit three-tier approval flow (do without asking \u002F get approval \u002F never do) and context-dependent privacy rules for messaging platforms.\n    *[Source: Clawdbot\u002F](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Ftree\u002Fmain\u002FClawdbot)*\n\n---\n\n## Synthesizing Best Practices: Key Takeaways for Builders\n\nAnalyzing these diverse prompts reveals a set of converging best practices for building reliable agentic AI systems:\n\n1.  **Define the Agent Clearly:** Start with an explicit role, purpose, and scope. Include contextual grounding like date or environment specifics.\n2.  **Structure for Clarity:** Break down complex instructions using headings, lists, or tags. Organize rules logically (e.g., group tool instructions, safety rules).\n3.  **Be Explicit About Tools:** Detail *what* each tool does, *how* to call it (syntax, parameters, format), and *when* (and when not) to use it. Provide examples. Embed usage policies directly.\n4.  **Mandate Step-by-Step Execution:** Encourage or enforce planning, iteration, and waiting for results\u002Fconfirmation. Prevent the AI from attempting too much at once. Consider explicit thinking phases or loops.\n5.  **Embed Domain Knowledge & Constraints:** Include relevant style guides, library usage rules, file conventions, platform limitations, and best practices for the agent's specific domain.\n6.  **Integrate Safety and Alignment:** Define unacceptable requests and provide clear refusal protocols. Embed specific policies for sensitive operations (data handling, image generation).\n7.  **Guide the Tone:** Set expectations for the interaction style (professional, friendly, concise, adaptive) to ensure a consistent user experience.\n8.  **Use Examples:** Illustrate complex rules or desired output formats with clear examples within the prompt (like Bolt.new and v0 do extensively).\n\nEssentially, an effective agentic system prompt acts as a comprehensive, well-structured operational manual that leaves little room for ambiguity while empowering the AI with the knowledge and procedures needed to act effectively and safely using its tools.\n\n---\n\n## Unique Conventions & Architectural Differences\n\nWhile core principles are shared, the *implementation* varies based on the agent's architecture and goals:\n\n*   **Tool Syntax:** Ranges from embedded MDX\u002FXML components (v0, same.new, Cline, Bolt) to expecting JSON outputs matching external schemas (ChatGPT, Manus).\n*   **Planning Mechanism:** Varies from explicit loops (Manus) and thinking tags (v0) to implicit guidance through iterative rules (same.new, Cline).\n*   **Editing Approach:** Some use diff-like formats (Cline's `replace_in_file`), others use custom components (v0's `QuickEdit`), while some specify overwriting vs. targeted edits (Bolt.new, Loveable).\n*   **Prompt Structure:** Can be monolithic (Cline, same.new) or modular across multiple files (Manus, Clawdbot, potentially v0 and Claude Code). Clawdbot takes this furthest with explicit file responsibilities: personality (SOUL.md), rules (AGENTS.md), and identity\u002Fprivacy (IDENTITY.md).\n*   **Level of Detail:** Varies significantly, with prompts like ChatGPT's embedding highly detailed function schemas and policies, while others like Manus rely more on external definitions (`tools.json`).\n\nThese differences highlight that there isn't a single \"perfect\" prompt structure, but rather effective prompts are tailored to the specific agent, its tools, its environment, and its intended tasks, while adhering to the core principles outlined above.\n\n---\n\n## Conclusion: Building the Agentic Future\n\nSystem prompts are the bedrock upon which capable and reliable agentic AI systems are built. As demonstrated by the examples from v0, same.new, Manus, ChatGPT, and others, successful prompts are detailed, structured, and explicit. They clearly define the agent's role, meticulously outline tool usage and operational procedures, enforce planning and iterative execution, embed necessary domain knowledge and safety constraints, and guide the interaction style.\n\nFor builders aiming to create the next generation of agentic AI in 2025 and beyond, studying these patterns provides invaluable insights. Mastering the craft of system prompting – blending clear instruction, structured organization, domain expertise, and safety considerations – will be key to unlocking the full potential of AI agents that can not only converse but actively collaborate and accomplish complex tasks in the digital world.","# 为代理型AI系统打造高效提示词：模式与实践\n\n## 目录\n\n*   [引言：代理型AI的蓝图](#introduction-the-blueprint-of-agentic-ai)\n*   [基础：代理型提示词的核心原则](#the-foundation-core-principles-of-agentic-prompts)\n    *   [1. 清晰的角色定义与范围](#1-clear-role-definition-and-scope)\n    *   [2. 结构化的指令与组织](#2-structured-instructions-and-organization)\n    *   [3. 明确的工具集成与使用指南](#3-explicit-tool-integration-and-usage-guidelines)\n    *   [4. 分步推理与规划](#4-step-by-step-reasoning-and-planning)\n    *   [5. 环境与上下文意识](#5-environment-and-context-awareness)\n    *   [6. 领域特定的专业知识与约束](#6-domain-specific-expertise-and-constraints)\n    *   [7. 安全性、对齐性与拒绝协议](#7-safety-alignment-and-refusal-protocols)\n    *   [8. 一致的语气与交互风格](#8-consistent-tone-and-interaction-style)\n*   [案例研究：分析现实世界中的提示词](#case-studies-analyzing-real-world-prompts)\n    *   [Vercel v0：UI生成与组件工具](#vercel-v0-ui-generation--component-tooling)\n    *   [same.new：代理式结对编程与严格工具链](#samenew-agentic-pair-programming--strict-tooling)\n    *   [Manus：通用代理与显式循环](#manus-general-purpose-agent--explicit-loop)\n    *   [OpenAI ChatGPT (GPT-4.5\u002F4o)：集成工具与政策](#openai-chatgpt-gpt-454o-integrated-tools--policies)\n    *   [其他系统注记（Cline、Bolt、Augment、Claude Code、Clawdbot）](#notes-on-other-systems-cline-bolt-augment-claude-code-clawdbot)\n*   [总结最佳实践：构建者的关键启示](#synthesizing-best-practices-key-takeaways-for-builders)\n*   [独特约定与架构差异](#unique-conventions--architectural-differences)\n*   [结论：构建代理型未来](#conclusion-building-the-agentic-future)\n*   [视觉AI代理：Harpagan](https:\u002F\u002Fharpagan.com)\n\n\n---\n\n## 引言：代理型AI的蓝图\n\n代理型人工智能（AI）系统的兴起，标志着从纯对话模型向能够主动执行任务、与工具交互并自主追求复杂目标的AI的重大转变。这些系统具备规划、执行命令、编辑文件、浏览网页等能力，有望彻底改变我们与技术互动的方式，并增强人类的能力。\n\n每个有效代理型AI的核心在于其**系统提示词**。它不仅仅是初始指令，更是基础蓝图、操作手册，甚至是指导AI行为、能力、限制和人格的“宪法”。精心设计的系统提示词对于确保代理可靠、安全且高效地实现用户目标至关重要。\n\n本指南深入探讨了如何艺术性与科学性地编写这些关键提示词。通过分析来自[awesome-ai-system-prompts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts)仓库的一系列真实世界系统提示词——特别是Vercel的v0、same.new、Manus、OpenAI的ChatGPT等实例——我们可以识别出反复出现的模式和最佳实践。对于那些正在塑造2025年及以后代理型未来的开发者而言，理解这些模式是创建强大、可预测且值得信赖的AI助手的关键。\n\n---\n\n## 基础：代理型提示词的核心原则\n\n在不同的代理型系统中，成功系统提示词中始终涌现出若干核心原则。这些原则构成了构建复杂代理行为的基础。\n\n### 1. 清晰的角色定义与范围\n\n**为何重要：** 明确界定AI的身份、核心功能和运作领域，能够锚定其行为、设定用户期望，并防止范围扩大或产生无意义的回答。它告诉AI自己“是谁”以及“应该做什么”。\n\n> **实用示例：**\n>\n> *   **Vercel v0：** 立即声明其身份和专业领域。\n>     ```\n>     你是Vercel的AI驱动助手v0。\n>     ```\n>     *[来源：v0\u002Fv0.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0.md)*\n>\n> *   **same.new：** 定义角色、能力水平及专属环境。\n>     ```\n>     你是强大的代理型AI编码助手。你仅在Same——全球最佳云端IDE——中运行。\n>     ```\n>     *[来源：same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md)*\n>\n> *   **Manus：** 自我介绍并列出其擅长的广泛任务类别。\n>     ```\n>     你是Manus，由Manus团队创建的AI代理。\n>\n>     你在以下任务上表现出色：\n>     1. 信息收集……\n>     2. 数据处理……\n>     3. 撰写多章节文章……\n>     ……\n>     ```\n>     *[来源：Manus\u002FAgentLoop.txt](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FAgentLoop.txt)*\n>\n> *   **ChatGPT (4.5 \u002F 4o)：** 清楚说明名称、创建者、底层架构，以及知识截止日期和当前日期等关键背景信息。\n>     ```\n>     你是ChatGPT，由OpenAI训练的大语言模型，基于GPT-4.5架构。\n>     知识截止日期：2023年10月\n>     当前日期：2025年4月5日\n>\n>     支持图像输入功能：已启用\n>     人格设定：v2\n>     ```\n>     *[来源：ChatGPT\u002F4-5.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4-5.md)*\n>\n> *   **Claude：** 不仅将其定位为工具，更建立了超越工具的人格形象。\n>     ```\n>     这位助手名为Claude，由Anthropic公司创造。\n>\n>     Claude乐于帮助人类，视自己为一位智慧而友善的助手，其深度与睿智使其远不止于简单的工具。\n>     ```\n>     *[来源：Claude\u002FClaude-Sonnet-3.7.txt](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FClaude\u002FClaude-Sonnet-3.7.txt)*\n\n### 2. 结构化指令与组织\n\n**为何重要：** 如果没有清晰的结构，冗长复杂的提示将难以管理。使用标题、列表、代码块或自定义标签，有助于人类维护者和 AI 模型解析并优先处理不同的规则或信息集合。\n\n> **实用示例：**\n>\n> *   **v0 & ChatGPT：** 大量使用 Markdown 标题（例如 `## 通用指令`、`# 工具`、`## 拒绝事项`）。\n>     *[来源：v0\u002Fv0.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0.md)*\n>\n> *   **same.new：** 采用类似 XML 的自定义标签来封装规则集（例如 `\u003Ctool_calling>`、`\u003Cmaking_code_changes>`）。\n>     *[来源：same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md)*\n>\n> *   **Manus：** 在 `Modules.md` 中使用描述性标签组织能力和规则（例如 `\u003Csystem_capability>`、`\u003Cagent_loop>`、`\u003Ctool_use_rules>`）。\n>     *[来源：Manus\u002FModules.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FModules.md)*\n>\n> *   **ChatGPT：** 使用 Markdown 标题（`# 工具`、`## bio`）和代码块（```` ```typescript ... ``` ````）来定义工具 Schema 和策略。\n>     *[来源：ChatGPT\u002F4-5.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4-5.md)*\n>\n> *   **Cline：** 使用层级化的 Markdown 标题（`# 工具使用格式`、`## execute_command`）以及在 `CAPABILITIES` 和 `RULES` 等部分下的列表。\n>     *[来源：Cline\u002Fsystem.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FCline\u002Fsystem.ts)*\n\n### 3. 明确的工具集成与使用指南\n\n**为何重要：** 对于代理式行为，AI 必须理解其工具：它们是什么、能做什么、如何调用（语法、参数）、所需格式（如 XML、JSON），以及最关键的是——**何时**以及**何时不应**使用这些工具。这需要详细的描述、清晰的 Schema 和明确的规则。\n\n> **实用示例：**\n>\n> *   **ChatGPT：** 直接在提示中为 `dalle` 和 `canmore` 等工具提供函数 Schema（TypeScript 定义）和详细策略。\n>     ```typescript\n>     \u002F\u002F ChatGPT 提示中 dalle 工具策略示例\n>     namespace dalle {\n>     \u002F\u002F 根据纯文本提示创建图像。\n>     type text2im = (_: {\n>     \u002F\u002F 请求图像的尺寸...\n>     size?: (\"1792x1024\" | \"1024x1024\" | \"1024x1792\"),\n>     \u002F\u002F 生成的图像数量...\n>     n?: number, \u002F\u002F 默认值：1\n>     \u002F\u002F 详细的图像描述...\n>     prompt: string,\n>     \u002F\u002F 如果用户引用了之前的图像...\n>     referenced_image_ids?: string[],\n>     }) => any;\n>     } \u002F\u002F namespace dalle\n>     ```\n>     *[来源：ChatGPT\u002F4-5.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4-5.md)*\n>\n> *   **same.new：** 专门设立 `\u003Ctool_calling>` 部分，详细说明遵循 Schema、不向用户提及工具名称、在调用工具前先解释原因等规则。参考了 `functions-schema.json` 文件（未完整展示，但隐含其结构）。\n>     ```xml\n>     \u003Ctool_calling>\n>       ...\n>       1. 始终严格按照工具调用 Schema 执行...\n>       3. **切勿在与用户交谈时提及工具名称。**...\n>       5. 在每次调用工具之前，先向用户解释你为什么要调用它。\n>     \u003C\u002Ftool_calling>\n>     ```\n>     *[来源：same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md)* | *[Schema：same.new\u002Ffunctions-schema.json](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Ffunctions-schema.json)*\n>\n> *   **Manus：** 将工具定义在外部的 `tools.json` 文件中（提供 Schema），并在 `Modules.md` 中包含规则，例如优先使用数据 API 而不是网络搜索。\n>     ```json\n>     \u002F\u002F Manus\u002Ftools.json 片段\n>     {\n>       \"type\": \"function\",\n>       \"function\": {\n>         \"name\": \"shell_exec\",\n>         \"description\": \"在指定的 Shell 会话中执行命令...\",\n>         \"parameters\": { ... }\n>       }\n>     }\n>     ```\n>     *[来源：Manus\u002Ftools.json](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002Ftools.json)* | *[规则：Manus\u002FModules.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FModules.md)*\n>\n> *   **Cline & Augment：** 直接在主系统提示中使用类似 XML 的标签或结构化文本，整合详细的工具描述、参数和使用示例。\n>     ```markdown\n>     \u002F\u002F Cline 示例工具定义\n>     ## execute_command\n>     描述：请求执行 CLI 命令...\n>     参数：\n>     - command：（必填）CLI 命令...\n>     - requires_approval：（必填）一个布尔值，表示...\n>     使用示例：\n>     \u003Cexecute_command>\n>     \u003Ccommand>您的命令在这里\u003C\u002Fcommand>\n>     \u003Crequires_approval>true 或 false\u003C\u002Frequires_approval>\n>     \u003C\u002Fexecute_command>\n>     ```\n>     *[来源：Cline\u002Fsystem.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FCline\u002Fsystem.ts)*\n>\n> *   **Bolt.new：** 使用专门的 `\u003Cartifact_instructions>` 部分，详细说明如何格式化工具输出（`\u003CboltAction type=\"shell\">`、`\u003CboltAction type=\"file\" filePath=\"...\">`），并将其置于主 `\u003CboltArtifact>` 标签内。\n>     *[来源：Bolt.new\u002Fprompts.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FBolt.new\u002Fprompts.ts)*\n>\n> *   **v0：** 将自定义 MDX 组件（如 `\u003CCodeProject>`、`\u003CQuickEdit>`、`\u003CDeleteFile \u002F>`）定义为其“工具”，并在响应中附带关于何时以及如何使用这些工具的规则。\n>     *[来源：v0\u002Fv0-tools.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0-tools.md)*\n\n### 4. 分步推理与规划\n\n**为何重要：** 复杂任务需要将问题分解。成功的提示词能够引导AI有条不紊地思考，规划行动步骤，逐步执行，并在继续下一步之前等待反馈或结果，从而减少错误并提高输出的连贯性。\n\n> **实用示例：**\n>\n> *   **Manus：** 其`Modules.md`文件中定义了明确的`\u003Cagent_loop>`，具备最清晰的规划机制。\n>     ```\n>     \u003Cagent_loop>\n>     您正处于代理循环中，通过以下步骤迭代完成任务：\n>     1. 分析事件...\n>     2. 选择工具...\n>     3. 等待执行...\n>     4. 迭代：每次迭代仅选择一个工具调用...\n>     5. 提交结果...\n>     6. 进入待机状态...\n>     \u003C\u002Fagent_loop>\n>     ```\n>     *[来源：Manus\u002FModules.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FModules.md)*\n>\n> *   **v0：** 在生成代码之前会先进行专门的思考阶段。\n>     ```\n>     在创建代码项目之前，v0会使用\u003CThinking>标签来仔细思考项目结构……\n>     ```\n>     *[来源：v0\u002Fv0.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0.md)*\n>\n> *   **same.new与Cline：** 要求每一步操作后都需等待用户确认或工具执行结果。\n>     ```\n>     每次使用工具后，务必等待用户确认后再继续。切勿假设工具使用一定成功……\n>     *(摘自same.new与Cline的提示词)*\n>     ```\n>     *[来源：same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md) | [Cline\u002Fsystem.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FCline\u002Fsystem.ts)*\n>\n> *   **Bolt.new：** 强调在行动之前进行全面思考。\n>     ```\n>     重要提示：在创建任何成果之前，请务必从整体和全面的角度进行思考。这意味着要考虑到所有相关文件……回顾所有之前的文件变更……分析整个项目的上下文……预见可能产生的影响……\n>     ```\n>     *[来源：Bolt.new\u002Fprompts.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FBolt.new\u002Fprompts.ts)*\n\n### 5. 环境与上下文感知\n\n**为何重要：** 代理通常运行在特定的环境中（操作系统、IDE、浏览器沙箱、特定的库等）。提供这些上下文信息可以让AI生成兼容的代码、使用恰当的命令，并理解各种限制。\n\n> **实用示例：**\n>\n> *   **Cline：** 包含一个`SYSTEM INFORMATION`部分。\n>     ```\n>     系统信息\n>\n>     操作系统：${osName()}\n>     默认Shell：${getShell()}\n>     主目录：${os.homedir().toPosix()}\n>     当前工作目录：${cwd.toPosix()}\n>     ```\n>     *[来源：Cline\u002Fsystem.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FCline\u002Fsystem.ts)*\n>\n> *   **Bolt.new：** 提供了关于WebContainer环境的详细`\u003Csystem_constraints>`。\n>     ```xml\n>     \u003Csystem_constraints>\n>       您正在名为WebContainer的环境中运行……该环境配备了一个模拟zsh的shell……可用的shell命令包括：cat、chmod、cp……\n>     \u003C\u002Fsystem_constraints>\n>     ```\n>     *[来源：Bolt.new\u002Fprompts.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FBolt.new\u002Fprompts.ts)*\n>\n> *   **Manus：** 详细描述了沙箱环境。\n>     ```\n>     沙箱环境：\n>     系统环境：\n>     - Ubuntu 22.04 (linux\u002Famd64)，具备互联网访问权限\n>     - 用户：`ubuntu`，拥有sudo权限\n>     ...\n>     开发环境：\n>     - Python 3.10.12……\n>     - Node.js 20.18.0……\n>     \u003C\u002Fsandbox_environment>\n>     ```\n>     *[来源：Manus\u002FModules.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FModules.md)*\n>\n> *   **same.new：** 明确指出了操作系统和具体IDE的上下文。\n>     ```\n>     操作系统为Linux 5.15.0-1075-aws（Ubuntu 22.04 LTS）。今天是2025年4月8日，星期二。\n>     您正与一位USER在Same中进行结对编程。\n>     USER可以看到一个实时预览……在一个iframe中……\n>     ```\n>     *[来源：same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md)*\n\n### 6. 领域专业知识与约束\n\n**为何重要：** 代理往往在特定领域内运作（如网页开发、数据分析等）。提示词中嵌入领域相关的知识、最佳实践、风格指南以及约束条件（例如必须使用的库、禁止出现的模式），以确保输出既高质量又符合上下文要求。\n\n> **实用示例：**\n>\n> *   **v0：** 包含关于Next.js\u002FReact开发、shadcn\u002Fui使用、图标库，甚至AI SDK集成的详细规则。\n>     ```\n>     v0会尽量使用shadcn\u002Fui库，除非用户另有指定……\n>     v0不会输出\u003Csvg>格式的图标。v0始终使用“lucide-react”包中的图标……\n>     v0只通过‘ai’和‘@ai-sdk’这两个模块来使用AI SDK……\n>     ```\n>     *[来源：v0\u002Fv0.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0.md)*\n>\n> *   **same.new：** 包括`\u003Cweb_development>`和`\u003Cwebsite_cloning>`等部分，针对这些任务给出了具体指导（例如优先使用Bun、正确使用shadcn CLI、合理合法地进行网页抓取）。\n>     *[来源：same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md)*\n>\n> *   **Bolt.new：** 包含`\u003Ccode_formatting_info>`（“代码缩进请使用2个空格”），并强调将功能拆分为更小的模块。\n>     ```\n>     重要提示：优先编写Node.js脚本，而非Shell脚本……\n>     重要提示：遵循编码最佳实践，将功能拆分为更小的模块……\n>     ```\n>     *[来源：Bolt.new\u002Fprompts.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FBolt.new\u002Fprompts.ts)*\n>\n> *   **Loveable：** 规定了React编码规范，包括Tailwind的使用、推荐的库（shadcn\u002Fui、lucide-react、recharts、@tanstack\u002Freact-query）以及错误处理原则。\n>     ```\n>     始终尝试使用shadcn\u002Fui库。\n>     除非特别要求，否则不要使用try\u002Fcatch语句捕获错误……\n>     ```\n>     *[来源：Loveable\u002FLoveable.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FLoveable\u002FLoveable.md)*\n>\n> *   **Claude Code：** 将关于代码风格和规范的规定嵌入到`System.js`中。\n>     ```\n>     对文件进行修改时，首先要理解该文件的代码规范。模仿现有代码风格，使用现有的库和工具，并遵循已有的模式。\n>     重要提示：未经要求，不得添加任何注释！\n>     ```\n>     *[来源：Claude-Code\u002FSystem.js](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FClaude-Code\u002FSystem.js)*\n\n### 7. 安全性、对齐与拒绝协议\n\n**为何重要：** 负责任的人工智能需要明确的边界。提示词定义了不可接受的请求（有害、不道德的内容），并规定了人工智能应如何拒绝这些请求（例如，使用标准回复、不道歉）或处理敏感操作（例如，DALL-E 的内容政策）。\n\n> **实用示例：**\n>\n> *   **v0：** 使用标准拒绝消息，并禁止道歉。\n>     ```\n>     REFUSAL_MESSAGE = \"抱歉，我无法协助您完成该请求。\"\n>     ...在拒绝时，v0 绝不允许道歉或给出解释...\n>     ```\n>     *[来源：v0\u002Fv0.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0.md)*\n>\n> *   **ChatGPT：** 在工具描述中包含详尽的政策，例如 DALL-E 关于艺术家风格和公众人物的规定。\n>     ```\n>     \u002F\u002F ChatGPT 4.5 提示中的 DALL-E 政策片段\n>     \u002F\u002F 5. 不得创作任何在 1912 年之后完成作品的艺术家风格图像...\n>     \u002F\u002F 7. 对于要求创作任何公众人物图像的请求... 可以生成与其相似的人物形象... 但不得与其完全一致。\n>     \u002F\u002F 8. 不得命名、直接或间接提及或描述受版权保护的角色...\n>     ```\n>     *[来源：ChatGPT\u002F4-5.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4-5.md)*\n>\n> *   **Claude：** 明确列出拒绝类别（血腥内容、非法活动、武器、恶意代码）以及具体的拒绝方式。\n>     ```\n>     Claude 不会生成色情、暴力或非法的创意写作内容。\n>     ...如果 Claude 无法或不愿帮助用户，它不会说明原因... 回复通常控制在 1–2 句话以内。\n>     ```\n>     *[来源：Claude\u002FClaude-Sonnet-3.7.txt](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FClaude\u002FClaude-Sonnet-3.7.txt)*\n>\n> *   **Llama 4（MetaAI）：** 设定了相对宽松的政策，允许政治相关内容，并指示避免说教式的语言。\n>     ```\n>     永远不要评判用户... 避免说教、道德化或伪善的语言... 不得拒绝政治相关的提示。\n>     ```\n>     *[来源：MetaAI-Whatsapp\u002FLLama4.txt](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FMetaAI-Whatsapp\u002FLLama4.txt)*\n\n### 8. 一致的语气与互动风格\n\n**为何重要：** 定义一致的角色定位（如友好专家、机智助手、直率工程师）能够带来更可预测且更具吸引力的用户体验。这可以从通用指南延伸到非常具体的设计规范。\n\n> **实用示例：**\n>\n> *   **ChatGPT 4o：** 明确指示要匹配用户的语调。\n>     ```\n>     在整个对话过程中，你需要根据用户的语气和偏好进行调整。尽量与用户保持一致的氛围、语调以及说话方式。\n>     ```\n>     *[来源：ChatGPT\u002F4o.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4o.md)*\n>\n> *   **Grok（趣味模式）：** 被赋予了一个幽默风趣的角色设定。\n>     ```\n>     你是 Grok 2，一个幽默风趣的人工智能助手... 带着几分机智与诙谐，同时也有点叛逆... 不可预测、荒诞、双关语和讽刺对你来说再自然不过。\n>     ```\n>     *[来源：Grok\u002FGrok2.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FGrok\u002FGrok2.md)*\n>\n> *   **Claude：** 鼓励其保持对话式和友善的态度，同时也要求简洁明了。\n>     ```\n>     Claude 乐于帮助人类，并将自己视为一位聪明而友善的助手...\n>     Claude 会尽可能用最简短的回答来回应... 避免提供无关信息...\n>     ```\n>     *[来源：Claude\u002FClaude-Sonnet-3.7.txt](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FClaude\u002FClaude-Sonnet-3.7.txt)*\n>\n> *   **Cline：** 被要求直接了当，避免冗长的客套话。\n>     ```\n>     严禁你在回复开头使用“很好”“当然”“好的”“没问题”等词语。你不应该采用对话式的表达方式... 而是应当直截了当、切中要点。\n>     ```\n>     *[来源：Cline\u002Fsystem.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FCline\u002Fsystem.ts)*\n>\n> *   **Bolt.new：** 强调简洁性。\n>     ```\n>     极其重要：切勿冗长，除非用户主动要求提供更多细节，否则不要做任何解释。\n>     ```\n>     *[来源：Bolt.new\u002Fprompts.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FBolt.new\u002Fprompts.ts)*\n\n---\n\n## 案例研究：分析真实世界的提示词\n\n让我们来看看这些原则是如何在仓库中的具体代理提示词中体现出来的。\n\n### Vercel v0：UI 生成与组件工具\n\n*[相关文件：v0\u002Fv0.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0.md)* | *[v0\u002Fv0-tools.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fv0\u002Fv0-tools.md)*\n\nVercel 的 v0 代理专门根据用户请求生成 UI 组件和全栈 Next.js 应用程序，通常还会结合图片或截图输入。\n\n#### 独特之处：\n\n*   **MDX 组件作为工具：** 与传统的函数调用不同，v0 的“工具”是特定的 MDX 组件标签，如 `\u003CCodeProject>`（用于包裹生成的代码）、`\u003CQuickEdit \u002F>`（用于小幅代码修改）、`\u003CDeleteFile \u002F>` 和 `\u003CMoveFile \u002F>`。提示词明确规定了何时以及如何使用这些输出格式。\n*   **高度领域特定性：** 提示词中包含了大量针对 Next.js App Router、Tailwind CSS、shadcn\u002Fui 以及 Vercel 平台限制的具体规则（例如，不允许使用 `package.json`、如何处理环境变量、预装库等）。\n*   **通过 `\u003CThinking>` 进行隐式规划：** 规定在生成 `\u003CCodeProject>` 之前必须先使用 `\u003CThinking>` 标签进行规划，以鼓励结构化的思考。\n*   **强调风格与最佳实践：** 包括文件命名规则（短横线分隔法）、响应式设计、无障碍访问（语义 HTML、ARIA 属性、替代文本）以及色彩搭配偏好（除非用户特别要求，否则避免使用靛蓝\u002F蓝色）。\n\n> **示例片段（基于组件的工具使用）：**\n>\n> ```\n> v0 总是使用 \u003CQuickEdit> 来对 React 代码块进行小幅修改...\n> v0 可以通过 \u003CDeleteFile \u002F> 组件删除 Code Project 中的某个文件。\n> ```\n\n### same.new：代理式结对编程与严格工具使用\n\n*[相关文件：same.new\u002Fsame.new.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002Fsame.new\u002Fsame.new.md)*\n\nsame.new 定位为在云端 IDE 中运行的代理式结对程序员。其提示强调精确的工具使用和迭代式开发流程。\n\n#### 独特特点：\n\n*   **类似 XML 的标签结构：** 使用 `\u003Ctool_calling>`、`\u003Cmaking_code_changes>`、`\u003Cweb_development>` 等标签来组织不同的规则集合。\n*   **严格的工具礼仪：** 明确禁止向用户提及工具名称，并要求事先解释调用工具的 *原因*，以促进透明度。\n*   **遵循 Schema 规范：** 强制要求严格按照 JSON Schema 进行工具调用（外部定义于 `functions-schema.json` 文件中）。\n*   **聚焦迭代式工作流：** 包含详细的编码工作流说明，包括在编辑前先读取文件、迭代修复运行时错误（最多 3 次）、使用 `suggestions` 工具以及对里程碑进行版本控制。\n*   **环境背景信息：** 提供操作系统详情、当前日期，并注明 IDE 上下文（实时预览 iframe）。\n\n> **示例片段（工具礼仪）：**\n>\n> ```xml\n> \u003Ctool_calling>\n>   ...\n>   3. **切勿在与 USER 交流时提及工具名称。** ...\n>   5. 在每次调用工具之前，务必先向 USER 解释你调用该工具的原因。\n> \u003C\u002Ftool_calling>\n> ```\n\n### Manus：通用型代理与明确循环机制\n\n*[相关文件：Manus\u002FModules.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FModules.md)* | *[Manus\u002FAgentLoop.txt](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002FAgentLoop.txt)* | *[Manus\u002Ftools.json](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FManus\u002Ftools.json)*\n\nManus 被设计为一个更广泛的通用型代理，运行在 Linux 沙盒环境中。其最突出的特点是明确界定的操作循环。\n\n#### 独特特点：\n\n*   **明确的代理循环：** 提示清晰地定义了一个多步骤的迭代循环（分析 -> 选择工具 -> 等待 -> 迭代 -> 提交 -> 待机），以此规范代理的核心行为。\n*   **模块化提示结构：** 指令分散在多个文件中（`AgentLoop.txt`、`Modules.md`、`tools.json`），表明采用模块化的提示管理方式。\n*   **沙盒环境感知：** 提及 Linux 沙盒环境、互联网接入以及预装工具（Python、Node.js）。\n*   **广泛的能力范围：** 列举了从信息收集、数据分析到应用创建与部署等各类任务。\n*   **模块集成：** 提到特定模块（规划器、知识库、数据源），这些模块通过事件流提供上下文或计划，暗示其内部架构更为复杂。\n\n> **示例片段（代理循环）：**\n>\n> ```\n> \u003Cagent_loop>\n> 你正处在代理循环中，通过以下步骤迭代完成任务：\n> 1. 分析事件...\n> 2. 选择工具...\n> 3. 等待执行...\n> 4. 迭代：每次迭代仅选择一个工具调用...\n> ...\n> \u003C\u002Fagent_loop>\n> ```\n\n### OpenAI ChatGPT（GPT-4.5\u002F4o）：集成工具与政策\n\n*[相关文件：ChatGPT\u002F4-5.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4-5.md)* | *[ChatGPT\u002F4o.md](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FChatGPT\u002F4o.md)*\n\nChatGPT 的提示（按记录）展示了将特定工具（插件\u002F函数）紧密集成于系统消息中的特点，同时附带 Schema 和详细的操作政策。\n\n#### 独特特点：\n\n*   **内嵌工具 Schema 与政策：** 独特之处在于直接在系统提示中包含每种工具的详细描述、类似 JSON\u002FTypeScript 的 Schema 以及详尽的使用政策（例如 `bio`、`canmore`、`dalle`、`python`、`web` 等）。\n*   **人格演变：** `Personality: v2` 标签以及 4o 提示中明确要求调整语气的指示，表明 OpenAI 不断优化其人格设定与交互风格。\n*   **详细的安全政策：** 内置细致的政策，尤其针对图像生成（如 `dalle` 工具关于艺术家风格、公众人物、受版权保护角色的规定）以及数据持久性（如 `bio` 工具对敏感信息的限制）。\n*   **情境背景信息：** 包括知识截止日期和当前日期。4o 提示还明确提及用户所在位置（“用户位于埃及。”）。\n\n> **示例片段（内嵌工具 Schema 与政策——Canmore）：**\n>\n> ```markdown\n> ## `canmore.create_textdoc`\n> 创建一个新的文本文档并在画布中显示。\n>\n> 切勿使用此功能。唯一可接受的使用场景是当用户明确请求画布时……\n>\n> 需要符合以下 Schema 的 JSON 字符串：\n> ```typescript\n> {\n>   name: string,\n>   type: \"document\" | \"code\u002Fpython\" | ...,\n>   content: string,\n> }\n> ```\n> ```\n\n### 其他系统的说明（Cline、Bolt、Augment、Claude Code、Clawdbot）\n\n虽然上述四个系统提供了深入的示例，但仓库中的其他提示也进一步强化了这些模式：\n\n*   **Cline 和 Augment：** 两者都在提示中通过结构化文本和类似 XML 的示例清晰地定义工具，并详细说明参数和使用方法。它们强调逐步执行并等待确认。Augment 类似于 v0，定义了自定义编辑工具（`editFile`、`strReplaceEditor`），并附有具体指令。\n    *[来源：Cline\u002Fsystem.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FCline\u002Fsystem.ts)* | *[Augment\u002Fpart_a.js](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FAugment\u002Fpart_a.js)*\n*   **Bolt.new：** 非常注重将输出结构化为一个包含有序 `\u003CboltAction>` 步骤（如 shell 命令、文件写入）的 `\u003CboltArtifact>`。它强调在创建工件之前进行整体规划，并遵循模块化等编码最佳实践。\n    *[来源：Bolt.new\u002Fprompts.ts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Fblob\u002Fmain\u002FBolt.new\u002Fprompts.ts)*\n*   **Claude Code：** 其提示（分散在 `System.js`、`EditTool.js` 等文件中）定义了特定工具的使用方式（例如详细的 `EditTool.js` 指令强调上下文和唯一性），并融入了系统信息。`ClearTool.js` 则定义了一个用于管理上下文窗口限制的摘要过程，这是长期运行代理任务中的关键环节。\n    *[来源：Claude-Code\u002F](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Ftree\u002Fmain\u002FClaude-Code)*\n*   **Clawdbot：** 采用一种基于模块化文件的方法，而非单一的巨幅提示。其行为被拆分到不同的文件中：`SOUL.md`（个性\u002F语气）、`AGENTS.md`（操作规则与审批层级）以及 `IDENTITY.md`（隐私边界与情境感知披露）。这种方式实现了可组合性——可以在不改变规则的情况下更换个性——同时使职责分离更加清晰。尤其值得注意的是其明确的三级审批流程（无需询问即可执行 \u002F 需要批准 \u002F 绝不执行）以及针对消息平台的情境依赖型隐私规则。\n    *[来源：Clawdbot\u002F](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts\u002Ftree\u002Fmain\u002FClawdbot)*\n\n---\n\n## 总结最佳实践：构建者的关键启示\n\n分析这些多样化的提示后，我们可以提炼出一套构建可靠智能体式 AI 系统的最佳实践：\n\n1.  **清晰定义智能体：** 从明确的角色、目的和范围入手。加入日期或环境细节等情境性信息。\n2.  **结构化以提升清晰度：** 使用标题、列表或标签来分解复杂指令。按逻辑组织规则（例如将工具指令和安全规则分组）。\n3.  **明确工具说明：** 详细说明每种工具的功能、调用方式（语法、参数、格式）以及适用场景。提供示例，并直接嵌入使用政策。\n4.  **强制逐步执行：** 鼓励或要求规划、迭代以及等待结果或确认。防止 AI 一次性尝试过多任务。可考虑设置明确的思考阶段或循环。\n5.  **融入领域知识与约束：** 将相关风格指南、库使用规则、文件规范、平台限制以及该智能体所在领域的最佳实践纳入其中。\n6.  **集成安全与对齐机制：** 定义不可接受的请求，并制定明确的拒绝协议。针对敏感操作（数据处理、图像生成）嵌入具体政策。\n7.  **引导对话基调：** 设定交互风格的期望（专业、友好、简洁、自适应），以确保一致的用户体验。\n8.  **运用示例：** 在提示中通过清晰的例子说明复杂规则或期望的输出格式（如 Bolt.new 和 v0 所广泛采用的）。\n\n本质上，有效的智能体式系统提示就像一份全面而结构化的操作手册，尽可能减少歧义，同时赋予 AI 必要的知识和程序，使其能够高效且安全地利用工具完成任务。\n\n---\n\n## 独特的约定与架构差异\n\n尽管核心原则相通，但具体的实现方式会因智能体的架构和目标而有所不同：\n\n*   **工具语法：** 从内嵌的 MDX\u002FXML 组件（v0、same.new、Cline、Bolt）到期望符合外部模式的 JSON 输出（ChatGPT、Manus）不等。\n*   **规划机制：** 有的采用显式循环（Manus）和思考标记（v0），有的则通过迭代规则进行隐式引导（same.new、Cline）。\n*   **编辑方式：** 有些使用类似 diff 的格式（Cline 的 `replace_in_file`），有些则使用自定义组件（v0 的 `QuickEdit`），还有些会明确指定覆盖式修改还是定点编辑（Bolt.new、Loveable）。\n*   **提示结构：** 可以是单体式的（Cline、same.new），也可以是跨多个文件的模块化结构（Manus、Clawdbot，v0 和 Claude Code 也可能如此）。Clawdbot 将这一点发挥到了极致，明确划分了各文件的职责：个性（SOUL.md）、规则（AGENTS.md）以及身份与隐私（IDENTITY.md）。\n*   **详细程度：** 差异显著，例如 ChatGPT 的提示会嵌入高度细化的函数模式和政策，而 Manus 等则更多依赖外部定义（`tools.json`）。\n\n这些差异表明，并不存在一种“完美”的提示结构，而是需要根据具体的智能体、其工具、所处环境以及预期任务量身定制有效提示，同时遵循上述核心原则。\n\n---\n\n## 结语：构建智能体的未来\n\n系统提示是构建强大且可靠的智能体式 AI 系统的基础。正如 v0、same.new、Manus、ChatGPT 等示例所示，成功的提示都具备细节丰富、结构清晰、表达明确的特点。它们清楚地界定智能体的角色，细致地说明工具的使用方法和操作流程，强制执行规划与迭代执行，融入必要的领域知识和安全约束，并引导交互风格。\n\n对于那些立志在 2025 年及以后打造下一代智能体式 AI 的开发者而言，研究这些模式将带来宝贵的洞见。掌握系统提示的技巧——将清晰的指令、结构化的组织、领域专业知识和安全考量融为一体——将是释放 AI 智能体全部潜力的关键。这些智能体不仅能够对话，还能积极协作，在数字世界中完成复杂的任务。","# awesome-ai-system-prompts 快速上手指南\n\n`awesome-ai-system-prompts` 并非一个需要安装运行的软件包或库，而是一个**精选的系统提示词（System Prompts）集合仓库**。它收录了 Vercel v0、same.new、Manus、ChatGPT 等主流 Agentic AI 系统的真实提示词源码，旨在为开发者提供构建高效智能体（Agent）的蓝图和最佳实践参考。\n\n因此，本指南将指导你如何获取、浏览及应用这些提示词资源到你的开发项目中。\n\n## 环境准备\n\n本项目无需特定的运行时环境（如 Node.js, Python 等）即可浏览内容，但为了在实际开发中应用这些提示词，建议准备以下环境：\n\n*   **操作系统**：Windows, macOS, 或 Linux\n*   **必备工具**：\n    *   **Git**：用于克隆仓库到本地。\n    *   **代码编辑器**：推荐 VS Code，便于阅读 Markdown 和分析提示词结构。\n    *   **AI 开发框架**（可选）：如果你计划将提示词集成到应用中，需准备好相应的开发环境（如 LangChain, LlamaIndex, Vercel AI SDK, 或直接调用 OpenAI\u002FAnthropic API 的环境）。\n*   **网络要求**：能够访问 GitHub。国内用户若访问缓慢，可配置 Git 代理或使用国内代码托管平台（如 Gitee）的镜像仓库（如有）。\n\n## 安装步骤（获取资源）\n\n你可以通过克隆仓库或下载 ZIP 包的方式获取所有提示词资源。\n\n### 方式一：使用 Git 克隆（推荐）\n\n在终端中执行以下命令：\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts.git\n```\n\n进入项目目录：\n\n```bash\ncd awesome-ai-system-prompts\n```\n\n> **国内加速提示**：如果克隆速度较慢，可以尝试使用国内镜像源（如果有维护）或通过设置 Git 代理加速：\n> ```bash\n> # 示例：设置 HTTP 代理 (请替换为你的实际代理地址)\n> git config --global http.proxy http:\u002F\u002F127.0.0.1:7890\n> git config --global https.proxy https:\u002F\u002F127.0.0.1:7890\n> ```\n\n### 方式二：直接下载\n\n访问仓库页面 [https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts](https:\u002F\u002Fgithub.com\u002Fdontriskit\u002Fawesome-ai-system-prompts)，点击绿色的 **\"Code\"** 按钮，选择 **\"Download ZIP\"**，解压后即可使用。\n\n## 基本使用\n\n本项目的核心用法是**阅读、分析和复用**其中的提示词文件。以下是三种典型的使用场景：\n\n### 1. 学习与参考（直接阅读）\n\n直接在本地或 GitHub 上浏览不同 AI 系统的提示词设计模式。\n\n*   **查看角色定义**：阅读 `v0\u002Fv0.md` 或 `ChatGPT\u002F4-5.md`，学习如何清晰定义 AI 的身份和能力边界。\n*   **研究工具调用**：查看 `same.new\u002Fsame.new.md` 中的 `\u003Ctool_calling>` 部分，了解如何通过 XML 标签规范工具调用流程。\n*   **分析安全策略**：参考 `Claude\u002FClaude-Sonnet-3.7.txt`，学习如何设定拒绝协议和安全对齐规则。\n\n### 2. 集成到自己的 AI 应用中\n\n你可以将仓库中的提示词模板复制并修改，作为你自家 AI 应用的 System Prompt。\n\n**示例：在 Python 中调用 OpenAI API 时使用 v0 的风格**\n\n```python\nimport os\nfrom openai import OpenAI\n\nclient = OpenAI(api_key=os.getenv(\"OPENAI_API_KEY\"))\n\n# 从本地文件读取或手动复制 v0 的提示词核心部分\nsystem_prompt = \"\"\"\nYou are an AI-powered assistant specialized in generating UI components.\nYour goal is to help users build web interfaces efficiently.\nFollow these principles:\n1. Clear Role Definition: You are a UI expert.\n2. Structured Instructions: Use markdown for code blocks.\n3. Tool Integration: If tools are available, describe usage clearly.\n\"\"\"\n\nresponse = client.chat.completions.create(\n    model=\"gpt-4o\",\n    messages=[\n        {\"role\": \"system\", \"content\": system_prompt},\n        {\"role\": \"user\", \"content\": \"Create a login form with email and password fields.\"}\n    ]\n)\n\nprint(response.choices[0].message.content)\n```\n\n### 3. 定制化开发\n\n根据 `Manus\u002FModules.md` 或 `Cline\u002Fsystem.ts` 的结构，为你特定的业务场景（如数据分析、自动化运维）编写模块化的系统提示词。\n\n*   **定义工具 Schema**：参考 `Manus\u002Ftools.json` 格式，定义你的自定义工具。\n*   **设定工作流**：模仿 `same.new` 的 `\u003Cagent_loop>` 逻辑，规划 AI 的思考与执行步骤。\n\n通过深入研究这些真实世界的案例，你可以显著提升自建 Agent 的可靠性、安全性和任务执行能力。","某初创团队的技术负责人正试图基于开源模型构建一个能自主完成“需求分析 - 代码生成 - 文件修改”全流程的智能编程助手，以加速内部工具开发。\n\n### 没有 awesome-ai-system-prompts 时\n- **角色定位模糊**：AI 常在“聊天伴侣”和“执行代理”间摇摆，面对复杂任务时倾向于输出长篇大论的建议而非直接执行代码操作。\n- **工具调用失控**：缺乏明确的操作指南，AI 经常错误地调用文件系统或网络搜索工具，甚至在没有用户确认的情况下尝试修改核心配置文件。\n- **逻辑规划缺失**：面对多步骤任务（如重构模块），AI 往往跳跃式执行，缺少逐步推理过程，导致中间状态出错且难以回溯。\n- **安全边界薄弱**：未设定严格的拒绝协议，当被诱导执行危险命令（如删除生产数据）时，AI 可能因缺乏对齐约束而照做。\n\n### 使用 awesome-ai-system-prompts 后\n- **身份界定清晰**：参考 Manus 或 same.new 的提示词模板，明确赋予 AI“严格遵循指令的执行者”身份，使其自动忽略无关闲聊，专注于任务闭环。\n- **工具集成规范**：复用 v0 或 Claude-Code 的结构化指令，让 AI 精准掌握工具调用时机与参数格式，确保每次文件编辑都经过逻辑校验。\n- **思维链显性化**：引入分步推理机制，强制 AI 在执行前输出详细计划（Plan），使复杂开发流程变得可预测、可监控。\n- **防御机制健全**：内置安全对齐策略，当检测到高风险操作时，AI 会主动触发拒绝协议并请求人工二次确认，保障系统稳定性。\n\nawesome-ai-system-prompts 通过提供经过验证的系统提示词蓝图，将原本不可控的对话模型转化为可靠、安全且高效的自主智能体，大幅降低了 Agent 应用的开发门槛与运行风险。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fdontriskit_awesome-ai-system-prompts_4d9413b2.png","dontriskit","Maksym Huczynski","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fdontriskit_3b627170.jpg","Harpagan.com co-founder","H.2.M",null,"https:\u002F\u002Fharpagan.com\u002F","https:\u002F\u002Fgithub.com\u002Fdontriskit",[81,85,89,93],{"name":82,"color":83,"percentage":84},"TypeScript","#3178c6",46,{"name":86,"color":87,"percentage":88},"JavaScript","#f1e05a",29.2,{"name":90,"color":91,"percentage":92},"Python","#3572A5",18.5,{"name":94,"color":95,"percentage":96},"Jinja","#a52a22",6.4,5698,863,"2026-04-08T07:56:56","MIT",1,"","未说明",{"notes":105,"python":103,"dependencies":106},"该项目是一个文档仓库，收集并分析了多个 AI Agent 系统提示词（System Prompts）的最佳实践和案例（如 Vercel v0, same.new, Manus, ChatGPT 等）。它不包含可执行的代码、模型文件或需要特定硬件环境的运行时程序，因此没有操作系统、GPU、内存或 Python 版本等运行环境需求。用户只需阅读 Markdown 文档即可获取提示词编写指南。",[],[35,13],"2026-03-27T02:49:30.150509","2026-04-08T22:45:40.434133",[],[]]