- About The Project
- 🚀 Key Features
- 🛠️ Getting Started
- 💡 Usage
- 📚 Definitive Guides & Protocols
- 🤝 Contributing
- 📄 License
- 📧 Contact
This repository is a dedicated resource for prompt engineers, developers, and AI enthusiasts. It provides a collection of structured protocols and guides for architecting sophisticated interactions with Large Language Models (LLMs). Our goal is to elevate the art and science of prompt engineering through expert-level examples, detailed guides, and reusable architectures.
- Expert-Level Protocols: A curated collection of advanced prompt engineering strategies.
- Definitive Guides: Includes comprehensive guides on system architecture and conversational dynamics for advanced prompt engineering.
- Reusable Architectures: Provides a master prompt for generating new, high-fidelity system and user prompts.
- Bilingual Documentation: Fully accessible in both English and Chinese.
Follow these simple steps to get a local copy up and running.
- A PDF reader for viewing the guides.
- A text editor for viewing the prompt files.
- Clone the repository:
git clone https://github.com/KuekHaoYang/AI-Prompt-Protocols
- Navigate to the project directory:
cd AI-Prompt-Protocols
The primary purpose of this repository is to serve as an educational and practical resource. You can:
- Study the Guides: Read the included PDF documents to deepen your understanding of advanced prompt engineering.
- Utilize the Protocols: Adapt the text-based prompt protocols for your own projects.
- Generate New Prompts: Use the
prompt_generation_prompt.txtwith a powerful LLM to architect new, specialized prompts for any task. - Explore KPrompt: Visit the KPrompt Website, a web-based application designed for generating high-quality prompts.
These resources provide the theoretical and practical foundation for advanced prompt engineering.
- Architecting Intelligence: A Definitive Guide to the Art and Science of Elite Prompt Engineering
- 构建智能：顶级提示词工程的艺术与科学权威指南
- Prompt Generation Protocol
graph TD;
subgraph book ["Architecting Intelligence: A Definitive Guide to Elite Prompt Engineering"];
direction LR;
%% Chapter 1: The Philosophy of Prompting
ch1("Ch 1: Philosophy of Prompting");
%% Chapter 2: Anatomy of an Effective Prompt
ch2("Ch 2: Anatomy of an Effective Prompt");
ch1 --> ch2;
subgraph sg_ch2 ["Anatomy"];
direction TB;
pillar_context["Pillar 1: Context ('Who' & 'Why')"];
pillar_task["Pillar 2: Task ('What')"];
pillar_format["Pillar 3: Format ('How')"];
ch2 --> pillar_context;
ch2 --> pillar_task;
ch2 --> pillar_format;
context_role["Assign a Role/Persona"];
context_bg["Provide Background Info"];
context_audience["Define the Audience"];
context_goal["State the Goal/Purpose"];
pillar_context --> context_role;
pillar_context --> context_bg;
pillar_context --> context_audience;
pillar_context --> context_goal;
task_verb["Use Strong, Unambiguous Verbs"];
task_chunking["Break Down Complexity (Chunking)"];
task_specific["Be Explicit and Specific"];
task_constraints["State Constraints and Boundaries"];
pillar_task --> task_verb;
pillar_task --> task_chunking;
pillar_task --> task_specific;
pillar_task --> task_constraints;
format_name["Explicitly Name the Format"];
format_fewshot["Provide Examples (Few-Shot Formatting)"];
format_structure["Specify Structural Elements & Tone"];
pillar_format --> format_name;
pillar_format --> format_fewshot;
pillar_format --> format_structure;
end
%% Chapter 3: System vs. User Prompts
ch3("Ch 3: The Two Pillars");
ch2 --> ch3;
subgraph sg_ch3 ["System vs. User Prompts"];
direction TB;
sys_prompt["System Prompt ('The Constitution / Character Brief')"];
usr_prompt["User Prompt ('The Conversational Directive')"];
ch3 --> sys_prompt;
ch3 --> usr_prompt;
sys_prop1["Defines AI's core identity, rules, constraints"];
sys_prop2["Persistent for the entire session"];
sys_prop3["Set pre-conversationally"];
sys_prompt --> sys_prop1;
sys_prompt --> sys_prop2;
sys_prompt --> sys_prop3;
usr_prop1["Specific, task-oriented instruction"];
usr_prop2["Dynamic, changes with each turn"];
usr_prop3["Action-driven and contextual"];
usr_prompt --> usr_prop1;
usr_prompt --> usr_prop2;
usr_prompt --> usr_prop3;
end
%% Foundational Strategies
strategies_foundation("Foundational Strategies");
ch3 --> strategies_foundation;
%% Chapter 4: Zero-Shot Strategy
ch4("Ch 4: The Zero-Shot Strategy");
strategies_foundation --> ch4;
ch4_desc["Instructing without providing examples"];
ch4 -- "Relies on model's intrinsic knowledge" --> ch4_desc;
%% Chapter 5: Few-Shot and One-Shot Strategy
ch5("Ch 5: The Few-Shot & One-Shot Strategy");
ch4 --> ch5;
ch5_desc["Guiding AI with 1+ examples (In-Context Learning)"];
ch5 -- "Used for specific formats, nuanced tasks, or unique styles" --> ch5_desc;
%% Chapter 6: The Persona Strategy
ch6("Ch 6: The Persona Strategy");
strategies_foundation --> ch6;
ch6_desc["Assigning an expert role to the AI"];
ch6_principle1["Principle 1: First-Person Identity ('I am...')"];
ch6_principle2["Principle 2: Elite Persona Instantiation"];
ch6_principle3["Principle 3: Archetypal Embodiment"];
ch6 --> ch6_desc;
ch6_desc --> ch6_principle1;
ch6_desc --> ch6_principle2;
ch6_desc --> ch6_principle3;
%% Content & Structure Strategies
strategies_content("Content & Structure Strategies");
strategies_foundation --> strategies_content;
%% Chapter 7: The Specificity Strategy
ch7("Ch 7: The Specificity Strategy");
strategies_content --> ch7;
ch7_desc["Eliminating ambiguity by providing details"];
ch7_tech1["Quantify everything possible"];
ch7_tech2["Define constraints and boundaries"];
ch7_tech3["Decompose into sub-tasks"];
ch7 --> ch7_desc;
ch7_desc --> ch7_tech1;
ch7_desc --> ch7_tech2;
ch7_desc --> ch7_tech3;
%% Chapter 8: The Contextual Priming Strategy
ch8("Ch 8: The Contextual Priming Strategy");
strategies_content --> ch8;
ch8_desc["Providing rich background to ground the AI"];
ch8_tech1["Provide source material (Grounding Data)"];
ch8_tech2["Provide situational context (The 'Why')"];
ch8_tech3["Leverage conversational history"];
ch8 --> ch8_desc;
ch8_desc --> ch8_tech1;
ch8_desc --> ch8_tech2;
ch8_desc --> ch8_tech3;
%% Chapter 9: The Structural Strategy
ch9("Ch 9: The Structural Strategy");
strategies_content --> ch9;
ch9_desc["Using delimiters and XML tags for clarity"];
ch9_tool1["Delimiters (###, ---) for simple separation"];
ch9_tool2["XML Tags (<tag>) for hierarchical control"];
ch9 --> ch9_desc;
ch9_desc --> ch9_tool1;
ch9_desc --> ch9_tool2;
%% Reasoning Strategies
strategies_reasoning("Advanced Reasoning Strategies");
strategies_content --> strategies_reasoning;
%% Chapter 10: Chain-of-Thought (CoT)
ch10("Ch 10: Chain-of-Thought (CoT) Strategy");
strategies_reasoning --> ch10;
ch10_desc["Instructing the model to 'think step-by-step'"];
ch10_zero["Zero-Shot CoT ('Let's think step by step.')"];
ch10_few["Few-Shot CoT (Providing examples of reasoning)"];
ch10 --> ch10_desc;
ch10_desc --> ch10_zero;
ch10_desc --> ch10_few;
%% Chapter 11: Self-Consistency
ch11("Ch 11: The Self-Consistency Strategy");
ch10 --> ch11;
ch11_desc["Enhancing CoT with multiple reasoning paths and majority voting"];
ch11_step1["1. Generate Diverse Paths (via Temperature)"];
ch11_step2["2. Extract Final Answer from Each Path"];
ch11_step3["3. Select Most Consistent Answer (Majority Vote)"];
ch11 --> ch11_desc;
ch11_desc --> ch11_step1;
ch11_step1 --> ch11_step2;
ch11_step2 --> ch11_step3;
%% Chapter 12: Tree-of-Thoughts (ToT)
ch12("Ch 12: The Tree-of-Thoughts (ToT) Strategy");
ch11 --> ch12;
ch12_desc["Exploring multiple solution branches simultaneously"];
ch12_step1["1. Thought Generation (Brainstorm multiple next steps)"];
ch12_step2["2. State Evaluation (Critique the viability of each step)"];
ch12_step3["3. Search & Pruning (Select best path, backtrack if needed)"];
ch12 --> ch12_desc;
ch12_desc --> ch12_step1;
ch12_step1 --> ch12_step2;
ch12_step2 --> ch12_step3;
%% Chapter 13: Step-Back Strategy
ch13("Ch 13: The Step-Back Strategy");
strategies_reasoning --> ch13;
ch13_desc["Generalizing a problem to unlock broader knowledge"];
ch13_stepA["Step 1: The Abstraction Prompt (Ask for general principles)"];
ch13_stepB["Step 2: The Application Prompt (Use principles as context for specific problem)"];
ch13 --> ch13_desc;
ch13_desc --> ch13_stepA;
ch13_stepA --> ch13_stepB;
%% Chapter 14: Self-Correction Strategy
ch14("Ch 14: The Self-Correction Strategy");
strategies_reasoning --> ch14;
ch14_desc["Prompting the AI to review and refine its own work"];
ch14_loop["Prompt -> [AI Draft] -> [AI Critique] -> [Final Output]"];
ch14 --> ch14_desc;
ch14_desc --> ch14_loop;
%% Agentic & Workflow Strategies
strategies_agentic("Agentic & Workflow Strategies");
strategies_reasoning --> strategies_agentic;
%% Chapter 15: ReAct Strategy
ch15("Ch 15: The ReAct Strategy (Reason + Act)");
strategies_agentic --> ch15;
ch15_desc["Combining reasoning with action via external tools"];
ch15_loop["Thought (Plan next action) --> Action (Use a tool) --> Observation (Process tool output)"];
ch15 --> ch15_desc;
ch15_desc --> ch15_loop;
ch15_loop --> ch15_loop;
%% Chapter 16: Prompt Chaining Strategy
ch16("Ch 16: The Prompt Chaining Strategy");
strategies_agentic --> ch16;
ch16_desc["Breaking down complex workflows into sequential, single-purpose prompts"];
ch16_flow["Prompt A -> Output A -> Input for Prompt B -> Output B"];
ch16 --> ch16_desc;
ch16_desc --> ch16_flow;
%% Chapter 17: Multi-Agent Strategy
ch17("Ch 17: The Multi-Agent Strategy");
ch16 --> ch17;
ch17_desc["Decomposing workflows into a team of specialized AI agents"];
ch17_team["Orchestrator --> Agent1(Persona A), Agent2(Persona B), Agent3(Persona C)"];
ch17 --> ch17_desc;
ch17_desc --> ch17_team;
%% Conversational & Refinement Strategies
strategies_refinement("Conversational & Refinement Strategies");
strategies_agentic --> strategies_refinement;
%% Chapter 18: Constructive Guidance
ch18("Ch 18: The Constructive Guidance Strategy");
strategies_refinement --> ch18;
ch18_desc["Iterating and steering the AI within a conversation"];
ch18_principle1["Use Positive Reinforcement"];
ch18_principle2["Manage State Explicitly"];
ch18_principle3["Correct Trajectory via Editing"];
ch18 --> ch18_desc;
ch18_desc --> ch18_principle1;
ch18_desc --> ch18_principle2;
ch18_desc --> ch18_principle3;
%% Chapter 19: Affirmative Direction
ch19("Ch 19: The Affirmative Direction Strategy");
strategies_refinement --> ch19;
ch19_desc["Stating what to do vs. what NOT to do"];
ch19_good["GOOD: 'Write in a warm, empathetic tone.'"];
ch19_bad["BAD: 'Do not sound like a robot.'"];
ch19 --> ch19_desc;
ch19_desc --> ch19_good;
ch19_desc --> ch19_bad;
%% Output & Parameter Control
strategies_control("Output & Parameter Control");
strategies_refinement --> strategies_control;
%% Chapter 20: Output Formatting
ch20("Ch 20: The Output Formatting Strategy");
strategies_control --> ch20;
ch20_desc["Forcing structured data like JSON and Tables"];
ch20_tech1["Direct Command"];
ch20_tech2["Provide Schema/Template"];
ch20_tech3["Few-Shot Example (Gold Standard)"];
ch20_tech4["Response Prefilling"];
ch20 --> ch20_desc;
ch20_desc --> ch20_tech1;
ch20_desc --> ch20_tech2;
ch20_desc --> ch20_tech3;
ch20_desc --> ch20_tech4;
%% Chapter 21: Response Prefilling
ch21("Ch 21: The Response Prefilling Strategy");
ch20 --> ch21;
ch21_desc["Seeding the assistant's answer for control"];
ch21_how["API Call: `{'role': 'assistant', 'content': '{'}`"];
ch21 --> ch21_desc;
ch21_desc --> ch21_how;
%% Chapter 22: Parameter Tuning
ch22("Ch 22: The Parameter Tuning Strategy");
strategies_control --> ch22;
ch22_desc["Engineering behavior with API parameters"];
ch22_temp["Temperature (Creativity/Randomness)"];
ch22_topk["Top-K (Whitelist of options)"];
ch22_topp["Top-P (Probability budget)"];
ch22 --> ch22_desc;
ch22_desc --> ch22_temp;
ch22_desc --> ch22_topk;
ch22_desc --> ch22_topp;
%% Specialized & Meta Strategies
strategies_meta("Specialized & Meta Strategies");
strategies_control --> strategies_meta;
%% Chapter 23: Long Context
ch23("Ch 23: The Long Context Strategy");
strategies_meta --> ch23;
ch23_desc["Optimizing for prompts with large data volumes"];
ch23_p1["Rule 1: Place Instructions Last"];
ch23_p2["Rule 2: Use Structural Tags to Index"];
ch23_p3["Rule 3: Force Active Retrieval Step"];
ch23 --> ch23_desc;
ch23_desc --> ch23_p1;
ch23_desc --> ch23_p2;
ch23_desc --> ch23_p3;
%% Chapter 24: Code Prompting
ch24("Ch 24: The Code Prompting Strategy");
strategies_meta --> ch24;
ch24_desc["Best practices for code tasks"];
ch24_gen["Generation (Provide full context & constraints)"];
ch24_dbg["Debugging (Provide full error & case file)"];
ch24_trans["Translation (Demand idiomatic adaptation)"];
ch24 --> ch24_desc;
ch24_desc --> ch24_gen;
ch24_desc --> ch24_dbg;
ch24_desc --> ch24_trans;
%% Chapter 25: Automatic Prompt Engineering (APE)
ch25("Ch 25: The Automatic Prompt Engineering (APE) Strategy");
strategies_meta --> ch25;
ch25_desc["Using AI to generate and optimize prompts"];
ch25_gen["1. Prompt Generation Phase"];
ch25_sel["2. Prompt Selection Phase"];
ch25 --> ch25_desc;
ch25_desc --> ch25_gen;
ch25_gen --> ch25_sel;
%% Chapter 26: Documentation Strategy
ch26("Ch 26: The Documentation Strategy");
strategies_meta --> ch26;
ch26_desc["The critical discipline of tracking and versioning prompts"];
ch26_why["Treat prompts like source code in version control (Git)"];
ch26 --> ch26_desc;
ch26_desc --> ch26_why;
%% Chapter 27: Unified Framework
ch27("Ch 27: Unified Framework");
strategies_meta --> ch27;
subgraph sg_ch27 ["Unified Framework"];
direction TB;
uf_pillar1["Pillar 1: Architecture"];
uf_pillar2["Pillar 2: Conversation"];
uf_pillar3["Pillar 3: Discipline"];
ch27 --> uf_pillar1;
ch27 --> uf_pillar2;
ch27 --> uf_pillar3;
uf_arch_decon["Deconstruct Workflow"];
uf_arch_pattern["Choose Pattern (Chain, Multi-Agent)"];
uf_arch_found["Define Foundation (Persona, Affirmative Dir.)"];
uf_pillar1 --> uf_arch_decon;
uf_pillar1 --> uf_arch_pattern;
uf_pillar1 --> uf_arch_found;
uf_conv_draft["First Output is a Draft"];
uf_conv_steer["Steer with Precision (Guidance)"];
uf_conv_reason["Synthesize Reasoning (CoT, ToT)"];
uf_pillar2 --> uf_conv_draft;
uf_pillar2 --> uf_conv_steer;
uf_pillar2 --> uf_conv_reason;
uf_disc_docs["Documentation Strategy"];
uf_disc_iter["Iteration & Evaluation Mindset"];
uf_pillar3 --> uf_disc_docs;
uf_pillar3 --> uf_disc_iter;
end
end
Contributions are what make the open-source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.
If you have a suggestion that would make this better, please fork the repo and create a pull request. You can also simply open an issue with the tag "enhancement".
Distributed under the MIT License. See LICENSE for more information.
Kuek Hao Yang - kuekhaoyang@gmail.com
Project Link: https://github.com/KuekHaoYang/AI-Prompt-Protocols
本仓库是为提示词工程师、开发者和AI爱好者准备的专属资源库。它提供了一系列结构化的协议和指南，用于构建与大型语言模型（LLM）的复杂交互。我们的目标是通过专家级的示例、详细的指南和可重用的架构，提升提示词工程的艺术与科学水平。
- 专家级协议： 精心策划的高级提示词工程策略集合。
- 权威指南： 包含关于高级提示词工程的系统架构和对话动态的综合指南。
- 可重用架构： 提供一个主提示词，用于生成新的、高保真的系统和用户提示词。
- 双语文档： 完全支持英文和中文。
请按照以下简单步骤在本地运行项目。
- 用于查看指南的PDF阅读器。
- 用于查看提示词文件的文本编辑器。
- 克隆仓库：
git clone https://github.com/KuekHaoYang/AI-Prompt-Protocols
- 进入项目目录：
cd AI-Prompt-Protocols
本仓库的主要目的是作为一个教育和实践资源。您可以：
- 学习指南： 阅读附带的PDF文档，加深对高级提示词工程的理解。
- 运用协议： 将基于文本的提示词协议应用于您自己的项目。
- 生成新提示词： 使用
prompt_generation_prompt.txt与强大的大型语言模型（LLM）结合，为任何任务构建新的、专门的提示词。 - 探索 KPrompt： 访问 KPrompt 网站，这是一个专门用于生成高质量提示词的网页应用程序。
这些资源为高级提示词工程提供了理论和实践基础。
- A Definitive Guide to Advanced Prompt Engineering: System Architecture and Conversational Dynamics
- 高级提示词工程权威指南：系统架构与对话动态
- 提示词生成协议
graph TD;
subgraph book ["《智能构建：高阶提示词工程权威指南》"];
direction LR;
%% Chapter 1: The Philosophy of Prompting
ch1("第一章：提示词的哲学");
%% Chapter 2: Anatomy of an Effective Prompt
ch2("第二章：高效提示词解析");
ch1 --> ch2;
subgraph sg_ch2 ["解析"];
direction TB;
pillar_context["第一支柱：语境（“谁”与“为何”）"];
pillar_task["第二支柱：任务（“何事”）"];
pillar_format["第三支柱：格式（“如何”）"];
ch2 --> pillar_context;
ch2 --> pillar_task;
ch2 --> pillar_format;
context_role["赋予角色/人设"];
context_bg["提供背景信息"];
context_audience["明确受众"];
context_goal["说明目标/目的"];
pillar_context --> context_role;
pillar_context --> context_bg;
pillar_context --> context_audience;
pillar_context --> context_goal;
task_verb["使用强劲、明确的动词"];
task_chunking["拆解复杂性（分块）"];
task_specific["明确具体"];
task_constraints["设定限制与边界"];
pillar_task --> task_verb;
pillar_task --> task_chunking;
pillar_task --> task_specific;
pillar_task --> task_constraints;
format_name["明确指定格式"];
format_fewshot["提供示例（小样本格式）"];
format_structure["指定结构要素与语气"];
pillar_format --> format_name;
pillar_format --> format_fewshot;
pillar_format --> format_structure;
end
%% Chapter 3: System vs. User Prompts
ch3("第三章：两大核心支柱");
ch2 --> ch3;
subgraph sg_ch3 ["系统提示词 vs. 用户提示词"];
direction TB;
sys_prompt["系统提示词（“宪法”/“角色设定”）"];
usr_prompt["用户提示词（“对话指令”）"];
ch3 --> sys_prompt;
ch3 --> usr_prompt;
sys_prop1["定义AI的核心身份、规则与约束"];
sys_prop2["贯穿整个会话"];
sys_prop3["在对话开始前设定"];
sys_prompt --> sys_prop1;
sys_prompt --> sys_prop2;
sys_prompt --> sys_prop3;
usr_prop1["具体、面向任务的指令"];
usr_prop2["动态性，随每次交互变化"];
usr_prop3["驱动行动并基于语境"];
usr_prompt --> usr_prop1;
usr_prompt --> usr_prop2;
usr_prompt --> usr_prop3;
end
%% Foundational Strategies
strategies_foundation("基础策略");
ch3 --> strategies_foundation;
%% Chapter 4: Zero-Shot Strategy
ch4("第四章：零样本策略");
strategies_foundation --> ch4;
ch4_desc["不提供示例进行指令"];
ch4 -- "依赖模型自身内在知识" --> ch4_desc;
%% Chapter 5: Few-Shot and One-Shot Strategy
ch5("第五章：小样本与单样本策略");
ch4 --> ch5;
ch5_desc["通过1个或多个示例引导AI（上下文学习）"];
ch5 -- "用于特定格式、细微任务或独特风格" --> ch5_desc;
%% Chapter 6: The Persona Strategy
ch6("第六章：人设策略");
strategies_foundation --> ch6;
ch6_desc["为AI赋予专家角色"];
ch6_principle1["原则一：第一人称身份（“我是一个…”）"];
ch6_principle2["原则二：精英人设具象化"];
ch6_principle3["原则三：原型化体现"];
ch6 --> ch6_desc;
ch6_desc --> ch6_principle1;
ch6_desc --> ch6_principle2;
ch6_desc --> ch6_principle3;
%% Content & Structure Strategies
strategies_content("内容与结构策略");
strategies_foundation --> strategies_content;
%% Chapter 7: The Specificity Strategy
ch7("第七章：明确性策略");
strategies_content --> ch7;
ch7_desc["通过提供细节消除歧义"];
ch7_tech1["尽可能量化所有内容"];
ch7_tech2["定义约束和边界"];
ch7_tech3["拆解为子任务"];
ch7 --> ch7_desc;
ch7_desc --> ch7_tech1;
ch7_desc --> ch7_tech2;
ch7_desc --> ch7_tech3;
%% Chapter 8: The Contextual Priming Strategy
ch8("第八章：语境铺垫策略");
strategies_content --> ch8;
ch8_desc["提供丰富背景以奠定AI基础"];
ch8_tech1["提供参考资料（基准数据）"];
ch8_tech2["提供情境语境（“为何”）"];
ch8_tech3["利用对话历史"];
ch8 --> ch8_desc;
ch8_desc --> ch8_tech1;
ch8_desc --> ch8_tech2;
ch8_desc --> ch8_tech3;
%% Chapter 9: The Structural Strategy
ch9("第九章：结构化策略");
strategies_content --> ch9;
ch9_desc["使用分隔符和XML标签提高清晰度"];
ch9_tool1["分隔符（###, ---）用于简单分隔"];
ch9_tool2["XML标签（<tag>）用于层级控制"];
ch9 --> ch9_desc;
ch9_desc --> ch9_tool1;
ch9_desc --> ch9_tool2;
%% Reasoning Strategies
strategies_reasoning("推理策略");
strategies_content --> strategies_reasoning;
%% Chapter 10: Chain-of-Thought (CoT)
ch10("第十章：思维链（CoT）策略");
strategies_reasoning --> ch10;
ch10_desc["指导模型“一步步思考”"];
ch10_zero["零样本思维链（“我们一步步思考。”）"];
ch10_few["小样本思维链（提供推理示例）"];
ch10 --> ch10_desc;
ch10_desc --> ch10_zero;
ch10_desc --> ch10_few;
%% Chapter 11: Self-Consistency
ch11("第十一章：自我一致性策略");
ch10 --> ch11;
ch11_desc["通过多条推理路径和多数投票增强思维链"];
ch11_step1["1. 生成多样路径（通过温度参数）"];
ch11_step2["2. 从每条路径中提取最终答案"];
ch11_step3["3. 选择最一致的答案（多数投票）"];
ch11 --> ch11_desc;
ch11_desc --> ch11_step1;
ch11_step1 --> ch11_step2;
ch11_step2 --> ch11_step3;
%% Chapter 12: Tree-of-Thoughts (ToT)
ch12("第十二章：思维树（ToT）策略");
ch11 --> ch12;
ch12_desc["同时探索多个解决方案分支"];
ch12_step1["1. 思维生成（集思广益多个下一步）"];
ch12_step2["2. 状态评估（批判性评估每一步的可行性）"];
ch12_step3["3. 搜索与剪枝（选择最佳路径，必要时回溯）"];
ch12 --> ch12_desc;
ch12_desc --> ch12_step1;
ch12_step1 --> ch12_step2;
ch12_step2 --> ch12_step3;
%% Chapter 13: Step-Back Strategy
ch13("第十三章：回溯策略");
strategies_reasoning --> ch13;
ch13_desc["泛化问题以解锁更广泛的知识"];
ch13_stepA["步骤一：抽象提示（要求给出一般原则）"];
ch13_stepB["步骤二：应用提示（将原则作为具体问题的语境）"];
ch13 --> ch13_desc;
ch13_desc --> ch13_stepA;
ch13_stepA --> ch13_stepB;
%% Chapter 14: Self-Correction Strategy
ch14("第十四章：自我纠正策略");
strategies_reasoning --> ch14;
ch14_desc["提示AI审查并完善自己的工作"];
ch14_loop["提示词 -> [AI草稿] -> [AI批判] -> [最终输出]"];
ch14 --> ch14_desc;
ch14_desc --> ch14_loop;
%% Agentic & Workflow Strategies
strategies_agentic("智能体与工作流策略");
strategies_reasoning --> strategies_agentic;
%% Chapter 15: ReAct Strategy
ch15("第十五章：ReAct策略（思考+行动）");
strategies_agentic --> ch15;
ch15_desc["通过外部工具结合推理与行动"];
ch15_loop["思考（规划下一步行动）--> 行动（使用工具）--> 观察（处理工具输出）"];
ch15 --> ch15_desc;
ch15_desc --> ch15_loop;
ch15_loop --> ch15_loop;
%% Chapter 16: Prompt Chaining Strategy
ch16("第十六章：提示词链策略");
strategies_agentic --> ch16;
ch16_desc["将复杂工作流拆解为按顺序、单一目的的提示词"];
ch16_flow["提示词A -> 输出A -> 提示词B的输入 -> 输出B"];
ch16 --> ch16_desc;
ch16_desc --> ch16_flow;
%% Chapter 17: Multi-Agent Strategy
ch17("第十七章：多智能体策略");
ch16 --> ch17;
ch17_desc["将工作流分解为由专业AI智能体组成的团队"];
ch17_team["协调器 --> 智能体1（人设A）、智能体2（人设B）、智能体3（人设C）"];
ch17 --> ch17_desc;
ch17_desc --> ch17_team;
%% Conversational & Refinement Strategies
strategies_refinement("对话与优化策略");
strategies_agentic --> strategies_refinement;
%% Chapter 18: Constructive Guidance
ch18("第十八章：建设性引导策略");
strategies_refinement --> ch18;
ch18_desc["在对话中迭代和引导AI"];
ch18_principle1["使用积极强化"];
ch18_principle2["明确管理状态"];
ch18_principle3["通过编辑修正轨迹"];
ch18 --> ch18_desc;
ch18_desc --> ch18_principle1;
ch18_desc --> ch18_principle2;
ch18_desc --> ch18_principle3;
%% Chapter 19: Affirmative Direction
ch19("第十九章：肯定性指示策略");
strategies_refinement --> ch19;
ch19_desc["说明做什么，而不是不做什么"];
ch19_good["正例：“以温暖、富有同理心的语气撰写。”"];
ch19_bad["反例：“不要听起来像个机器人。”"];
ch19 --> ch19_desc;
ch19_desc --> ch19_good;
ch19_desc --> ch19_bad;
%% Output & Parameter Control
strategies_control("输出与参数控制");
strategies_refinement --> strategies_control;
%% Chapter 20: The Output Formatting Strategy
ch20("第二十章：输出格式化策略");
strategies_control --> ch20;
ch20_desc["强制输出JSON和表格等结构化数据"];
ch20_tech1["直接指令"];
ch20_tech2["提供模式/模板"];
ch20_tech3["小样本示例（黄金标准）"];
ch20_tech4["预填充响应"];
ch20 --> ch20_desc;
ch20_desc --> ch20_tech1;
ch20_desc --> ch20_tech2;
ch20_desc --> ch20_tech3;
ch20_desc --> ch20_tech4;
%% Chapter 21: Response Prefilling
ch21("第二十一章：响应预填充策略");
ch20 --> ch21;
ch21_desc["预设助手答案以进行控制"];
ch21_how["API调用：`{'role': 'assistant', 'content': '{'}`"];
ch21 --> ch21_desc;
ch21_desc --> ch21_how;
%% Chapter 22: Parameter Tuning
ch22("第二十二章：参数调优策略");
strategies_control --> ch22;
ch22_desc["通过API参数调控行为"];
ch22_temp["Temperature（创造力/随机性）"];
ch22_topk["Top-K（选项白名单）"];
ch22_topp["Top-P（概率预算）"];
ch22 --> ch22_desc;
ch22_desc --> ch22_temp;
ch22_desc --> ch22_topk;
ch22_desc --> ch22_topp;
%% Specialized & Meta Strategies
strategies_meta("专项与元策略");
strategies_control --> strategies_meta;
%% Chapter 23: Long Context
ch23("第二十三章：长语境策略");
strategies_meta --> ch23;
ch23_desc["优化处理大数据量的提示词"];
ch23_p1["原则一：指令置于最后"];
ch23_p2["原则二：使用结构化标签进行索引"];
ch23_p3["原则三：强制主动检索步骤"];
ch23 --> ch23_desc;
ch23_desc --> ch23_p1;
ch23_desc --> ch23_p2;
ch23_desc --> ch23_p3;
%% Chapter 24: Code Prompting
ch24("第二十四章：代码提示词策略");
strategies_meta --> ch24;
ch24_desc["代码任务的最佳实践"];
ch24_gen["生成（提供完整上下文与约束）"];
ch24_dbg["调试（提供完整错误信息与案例文件）"];
ch24_trans["翻译（要求地道化适应）"];
ch24 --> ch24_desc;
ch24_desc --> ch24_gen;
ch24_desc --> ch24_dbg;
ch24_desc --> ch24_trans;
%% Chapter 25: Automatic Prompt Engineering (APE)
ch25("第二十五章：自动化提示词工程（APE）策略");
strategies_meta --> ch25;
ch25_desc["利用AI生成和优化提示词"];
ch25_gen["1. 提示词生成阶段"];
ch25_sel["2. 提示词选择阶段"];
ch25 --> ch25_desc;
ch25_desc --> ch25_gen;
ch25_gen --> ch25_sel;
%% Chapter 26: Documentation Strategy
ch26("第二十六章：文档化策略");
strategies_meta --> ch26;
ch26_desc["追踪和版本化提示词的关键纪律"];
ch26_why["将提示词视为版本控制（Git）中的源代码"];
ch26 --> ch26_desc;
ch26_desc --> ch26_why;
%% Chapter 27: Unified Framework
ch27("第二十七章：统一框架");
strategies_meta --> ch27;
subgraph sg_ch27 ["统一框架"];
direction TB;
uf_pillar1["第一支柱：架构"];
uf_pillar2["第二支柱：对话"];
uf_pillar3["第三支柱：规范"];
ch27 --> uf_pillar1;
ch27 --> uf_pillar2;
ch27 --> uf_pillar3;
uf_arch_decon["拆解工作流"];
uf_arch_pattern["选择模式（链式、多智能体）"];
uf_arch_found["定义基础（人设、肯定性指示）"];
uf_pillar1 --> uf_arch_decon;
uf_pillar1 --> uf_arch_pattern;
uf_pillar1 --> uf_arch_found;
uf_conv_draft["首次输出为草稿"];
uf_conv_steer["精准引导（指导）"];
uf_conv_reason["综合推理（思维链、思维树）"];
uf_pillar2 --> uf_conv_draft;
uf_pillar2 --> uf_conv_steer;
uf_pillar2 --> uf_conv_reason;
uf_disc_docs["文档化策略"];
uf_disc_iter["迭代与评估思维"];
uf_pillar3 --> uf_disc_docs;
uf_pillar3 --> uf_disc_iter;
end
end
贡献使开源社区成为一个学习、启发和创造的绝佳场所。我们非常感谢您的任何贡献。
如果您有任何建议能让这个项目变得更好，请 fork 本仓库并创建一个 pull request。您也可以简单地提交一个带有“enhancement”标签的 issue。
根据 MIT 许可证分发。更多信息请参见 LICENSE 文件。
Kuek Hao Yang - kuekhaoyang@gmail.com