ai-prompt-engineering
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChinesePrompt Engineering — Operational Skill
提示词工程 — 工程化实践技能
Modern Best Practices (January 2026): versioned prompts, explicit output contracts, regression tests, and safety threat modeling for tool/RAG prompts (OWASP LLM Top 10: https://owasp.org/www-project-top-10-for-large-language-model-applications/).
This skill provides operational guidance for building production-ready prompts across standard tasks, RAG workflows, agent orchestration, structured outputs, hidden reasoning, and multi-step planning.
All content is operational, not theoretical. Focus on patterns, checklists, and copy-paste templates.
2026年1月现代最佳实践:版本化提示词、明确的输出契约、回归测试,以及针对工具/RAG提示词的安全威胁建模(OWASP LLM十大风险:https://owasp.org/www-project-top-10-for-large-language-model-applications/)。
本技能为构建生产就绪型提示词提供工程化指导,覆盖标准任务、RAG工作流、Agent编排、结构化输出、隐藏推理以及多步骤规划等场景。
所有内容均为工程化实践,而非理论知识。重点关注模式、检查清单和可直接复制的模板。
Quick Start (60 seconds)
快速入门(60秒)
- Pick a pattern from the decision tree (structured output, extractor, RAG, tools/agent, rewrite, classification).
- Start from a template in and fill in
assets/,TASK,INPUT, andRULES.OUTPUT FORMAT - Add guardrails: instruction/data separation, “no invented details”, missing → /explicit missing.
null - Add validation: JSON parse check, schema check, citations check, post-tool checks.
- Add evals: 10–20 cases while iterating, 50–200 before release, plus adversarial injection cases.
- 从决策树中选择合适的模式(结构化输出、提取器、RAG、工具/Agent、重写、分类)。
- 从目录下的模板开始,填充
assets/、TASK、INPUT和RULES字段。OUTPUT FORMAT - 添加防护措施:指令与数据分离、“不得编造细节”、缺失信息返回/明确标注缺失。
null - 添加验证:JSON解析检查、Schema校验、引用检查、工具调用后检查。
- 添加评估:迭代过程中准备10-20个测试用例,发布前准备50-200个测试用例,同时包含对抗性注入测试用例。
Model Notes (2026)
模型注意事项(2026年)
This skill includes Claude Code + Codex CLI optimizations:
- Action directives: Frame for implementation, not suggestions
- Parallel tool execution: Independent tool calls can run simultaneously
- Long-horizon task management: State tracking, incremental progress, context compaction resilience
- Positive framing: Describe desired behavior rather than prohibitions
- Style matching: Prompt formatting influences output style
- Domain-specific patterns: Specialized guidance for frontend, research, and agentic coding
- Style-adversarial resilience: Stress-test refusals with poetic/role-play rewrites; normalize or decline stylized harmful asks before tool use
Prefer “brief justification” over requesting chain-of-thought. When using private reasoning patterns, instruct: think internally; output only the final answer.
本技能包含针对Claude Code + Codex CLI的优化建议:
- 行动指令:以实现为目标构建指令,而非仅给出建议
- 并行工具执行:独立的工具调用可同时运行
- 长周期任务管理:状态跟踪、增量进度、上下文压缩韧性
- 正向表述:描述期望的行为而非禁止的行为
- 风格匹配:提示词格式会影响输出风格
- 领域特定模式:针对前端、研究和Agent化编码的专业指导
- 风格对抗韧性:通过诗意/角色扮演式重写进行拒绝策略压力测试;在工具调用前规范化或拒绝风格化的有害请求
优先使用“简短理由说明”而非请求思维链。使用私有推理模式时,需指令:仅在内部思考;仅输出最终答案。
Quick Reference
快速参考
| Task | Pattern to Use | Key Components | When to Use |
|---|---|---|---|
| Machine-parseable output | Structured Output | JSON schema, "JSON-only" directive, no prose | API integrations, data extraction |
| Field extraction | Deterministic Extractor | Exact schema, missing->null, no transformations | Form data, invoice parsing |
| Use retrieved context | RAG Workflow | Context relevance check, chunk citations, explicit missing info | Knowledge bases, documentation search |
| Internal reasoning | Hidden Chain-of-Thought | Internal reasoning, final answer only | Classification, complex decisions |
| Tool-using agent | Tool/Agent Planner | Plan-then-act, one tool per turn | Multi-step workflows, API calls |
| Text transformation | Rewrite + Constrain | Style rules, meaning preservation, format spec | Content adaptation, summarization |
| Classification | Decision Tree | Ordered branches, mutually exclusive, JSON result | Routing, categorization, triage |
| 任务 | 适用模式 | 核心组件 | 使用场景 |
|---|---|---|---|
| 机器可解析输出 | 结构化输出 | JSON Schema、“仅输出JSON”指令、无散文式内容 | API集成、数据提取 |
| 字段提取 | 确定性提取器 | 精确Schema、缺失值返回null、无转换操作 | 表单数据、发票解析 |
| 使用检索到的上下文 | RAG工作流 | 上下文相关性检查、块引用、明确标注缺失信息 | 知识库、文档搜索 |
| 内部推理 | 隐藏思维链 | 内部推理、仅输出最终答案 | 分类、复杂决策 |
| 使用工具的Agent | 工具/Agent规划器 | 先规划后执行、每轮调用一个工具 | 多步骤工作流、API调用 |
| 文本转换 | 重写与约束 | 风格规则、语义保留、格式规范 | 内容适配、摘要生成 |
| 分类 | 决策树 | 有序分支、互斥规则、JSON结果 | 路由、分类、分流 |
Decision Tree: Choosing the Right Pattern
决策树:选择合适的模式
text
User needs: [Prompt Type]
|-- Output must be machine-readable?
| |-- Extract specific fields only? -> **Deterministic Extractor Pattern**
| `-- Generate structured data? -> **Structured Output Pattern (JSON)**
|
|-- Use external knowledge?
| `-- Retrieved context must be cited? -> **RAG Workflow Pattern**
|
|-- Requires reasoning but hide process?
| `-- Classification or decision task? -> **Hidden Chain-of-Thought Pattern**
|
|-- Needs to call external tools/APIs?
| `-- Multi-step workflow? -> **Tool/Agent Planner Pattern**
|
|-- Transform existing text?
| `-- Style/format constraints? -> **Rewrite + Constrain Pattern**
|
`-- Classify or route to categories?
`-- Mutually exclusive rules? -> **Decision Tree Pattern**text
用户需求: [提示词类型]
|-- 输出必须是机器可读的?
| |-- 仅提取特定字段? -> **确定性提取器模式**
| `-- 生成结构化数据? -> **结构化输出模式(JSON)**
|
|-- 需要使用外部知识?
| `-- 检索到的上下文必须被引用? -> **RAG工作流模式**
|
|-- 需要推理但隐藏过程?
| `-- 分类或决策任务? -> **隐藏思维链模式**
|
|-- 需要调用外部工具/API?
| `-- 多步骤工作流? -> **工具/Agent规划器模式**
|
|-- 转换现有文本?
| `-- 有风格/格式约束? -> **重写与约束模式**
|
`-- 分类或路由到不同类别?
`-- 规则互斥? -> **决策树模式**Copy/Paste: Minimal Prompt Skeletons
可直接复制的最简提示词骨架
1) Generic "output contract" skeleton
1) 通用“输出契约”骨架
text
TASK:
{{one_sentence_task}}
INPUT:
{{input_data}}
RULES:
- Follow TASK exactly.
- Use only INPUT (and tool outputs if tools are allowed).
- No invented details. Missing required info -> say what is missing.
- Keep reasoning hidden.
- Follow OUTPUT FORMAT exactly.
OUTPUT FORMAT:
{{schema_or_format_spec}}text
TASK:
{{one_sentence_task}}
INPUT:
{{input_data}}
RULES:
- 严格遵循TASK要求。
- 仅使用INPUT(若允许使用工具则包含工具输出)。
- 不得编造细节。缺失必要信息时,说明缺失内容。
- 隐藏推理过程。
- 严格遵循OUTPUT FORMAT要求。
OUTPUT FORMAT:
{{schema_or_format_spec}}2) Tool/agent skeleton (deterministic)
2) 确定性工具/Agent骨架
text
AVAILABLE TOOLS:
{{tool_signatures_or_names}}
WORKFLOW:
- Make a short plan.
- Call tools only when required to complete the task.
- Validate tool outputs before using them.
- If the environment supports parallel tool calls, run independent calls in parallel.text
可用工具:
{{tool_signatures_or_names}}
工作流:
- 制定简短计划。
- 仅在完成任务必需时调用工具。
- 使用工具输出前先进行验证。
- 若环境支持并行工具调用,可同时运行独立的调用请求。3) RAG skeleton (grounded)
3) 基于RAG的骨架(落地版)
text
RETRIEVED CONTEXT:
{{chunks_with_ids}}
RULES:
- Use only retrieved context for factual claims.
- Cite chunk ids for each claim.
- If evidence is missing, say what is missing.text
检索到的上下文:
{{chunks_with_ids}}
规则:
- 事实性声明仅使用检索到的上下文。
- 每个声明需引用块ID。
- 若缺少证据,说明缺失内容。Operational Checklists
工程化检查清单
Use these references when validating or debugging prompts:
frameworks/shared-skills/skills/ai-prompt-engineering/references/quality-checklists.mdframeworks/shared-skills/skills/ai-prompt-engineering/references/production-guidelines.md
验证或调试提示词时可参考以下文档:
frameworks/shared-skills/skills/ai-prompt-engineering/references/quality-checklists.mdframeworks/shared-skills/skills/ai-prompt-engineering/references/production-guidelines.md
Context Engineering (2026)
上下文工程(2026年)
True expertise in prompting extends beyond writing instructions to shaping the entire context in which the model operates. Context engineering encompasses:
- Conversation history: What prior turns inform the current response
- Retrieved context (RAG): External knowledge injected into the prompt
- Structured inputs: JSON schemas, system/user message separation
- Tool outputs: Results from previous tool calls that shape next steps
真正的提示词专业能力不仅限于编写指令,还包括塑造模型运行的整个上下文环境。上下文工程涵盖:
- 对话历史:哪些之前的对话回合会影响当前响应
- 检索到的上下文(RAG):注入到提示词中的外部知识
- 结构化输入:JSON Schema、系统/用户消息分离
- 工具输出:之前工具调用的结果会影响后续步骤
Context Engineering vs Prompt Engineering
上下文工程 vs 提示词工程
| Aspect | Prompt Engineering | Context Engineering |
|---|---|---|
| Focus | Instruction text | Full input pipeline |
| Scope | Single prompt | RAG + history + tools |
| Optimization | Word choice, structure | Information architecture |
| Goal | Clear instructions | Optimal context window |
| 维度 | 提示词工程 | 上下文工程 |
|---|---|---|
| 关注点 | 指令文本 | 完整输入管道 |
| 范围 | 单个提示词 | RAG + 历史 + 工具 |
| 优化方向 | 措辞、结构 | 信息架构 |
| 目标 | 清晰的指令 | 最优上下文窗口 |
Key Context Engineering Patterns
核心上下文工程模式
1. Context Prioritization: Place most relevant information first; models attend more strongly to early context.
2. Context Compression: Summarize history, truncate tool outputs, select most relevant RAG chunks.
3. Context Separation: Use clear delimiters (, , ) to separate instruction types.
<system><user><context>4. Dynamic Context: Adjust context based on task complexity - simple tasks need less context, complex tasks need more.
1. 上下文优先级:将最相关的信息放在最前面;模型对早期上下文的关注度更高。
2. 上下文压缩:总结历史对话、截断工具输出、选择最相关的RAG块。
3. 上下文分离:使用明确的分隔符(、、)区分不同类型的指令。
<system><user><context>4. 动态上下文:根据任务复杂度调整上下文——简单任务需要更少上下文,复杂任务需要更多。
Core Concepts vs Implementation Practices
核心概念 vs 实现实践
Core Concepts (Vendor-Agnostic)
核心概念(与厂商无关)
- Prompt contract: inputs, allowed tools, output schema, max tokens, and refusal rules.
- Context engineering: conversation history, RAG context, tool outputs, and structured inputs shape model behavior.
- Determinism controls: temperature/top_p, constrained decoding/structured outputs, and strict formatting.
- Cost & latency budgets: prompt length and max output drive tokens and tail latency; enforce hard limits and measure p95/p99.
- Evaluation: golden sets + regression gates + A/B + post-deploy monitoring.
- Security: prompt injection, data exfiltration, and tool misuse are primary threats (OWASP LLM Top 10: https://owasp.org/www-project-top-10-for-large-language-model-applications/).
- 提示词契约:输入、允许使用的工具、输出Schema、最大Token数和拒绝规则。
- 上下文工程:对话历史、RAG上下文、工具输出和结构化输入共同塑造模型行为。
- 确定性控制:temperature/top_p、约束解码/结构化输出、严格格式要求。
- 成本与延迟预算:提示词长度和最大输出长度决定Token消耗和尾部延迟;需强制执行硬限制并测量p95/p99指标。
- 评估:黄金测试集 + 回归门控 + A/B测试 + 部署后监控。
- 安全:提示词注入、数据泄露和工具滥用是主要威胁(OWASP LLM十大风险:https://owasp.org/www-project-top-10-for-large-language-model-applications/)。
Implementation Practices (Model/Platform-Specific)
实现实践(与模型/平台相关)
- Use model-specific structured output features when available; keep a schema validator as the source of truth.
- Align tracing/metrics with OpenTelemetry GenAI semantic conventions (https://opentelemetry.io/docs/specs/semconv/gen-ai/).
- 若模型支持特定的结构化输出功能则优先使用;将Schema验证器作为事实来源。
- 使追踪/指标与OpenTelemetry GenAI语义规范对齐(https://opentelemetry.io/docs/specs/semconv/gen-ai/)。
Do / Avoid
建议/禁忌
Do
- Do keep prompts small and modular; centralize shared fragments (policies, schemas, style).
- Do add a prompt eval harness and block merges on regressions.
- Do prefer "brief justification" over requesting chain-of-thought; treat hidden reasoning as model-internal.
Avoid
- Avoid prompt sprawl (many near-duplicates with no owner or tests).
- Avoid brittle multi-step chains without intermediate validation.
- Avoid mixing policy and product copy in the same prompt (harder to audit and update).
建议
- 保持提示词精简且模块化;集中管理共享片段(策略、Schema、风格)。
- 添加提示词评估框架,出现回归时阻止合并。
- 优先使用“简短理由说明”而非请求思维链;将隐藏推理视为模型内部行为。
禁忌
- 避免提示词泛滥(大量近乎重复的提示词,无所有者或测试)。
- 避免脆弱的多步骤链,无中间验证环节。
- 避免在同一个提示词中混合策略和产品文案(难以审计和更新)。
Navigation: Core Patterns
导航:核心模式
- Core Patterns - 7 production-grade prompt patterns
- Structured Output (JSON), Deterministic Extractor, RAG Workflow
- Hidden Chain-of-Thought, Tool/Agent Planner, Rewrite + Constrain, Decision Tree
- Each pattern includes structure template and validation checklist
- 核心模式 - 7种生产级提示词模式
- 结构化输出(JSON)、确定性提取器、RAG工作流
- 隐藏思维链、工具/Agent规划器、重写与约束、决策树
- 每种模式均包含结构模板和验证检查清单
Navigation: Best Practices
导航:最佳实践
-
Best Practices (Core) - Foundation rules for production-grade prompts
- System instruction design, output contract specification, action directives
- Context handling, error recovery, positive framing, style matching, style-adversarial red teaming
- Anti-patterns, Claude 4+ specific optimizations
-
Production Guidelines - Deployment and operational guidance
- Evaluation & testing (Prompt CI/CD), model parameters, few-shot selection
- Safety & guardrails, conversation memory, context compaction resilience
- Answer engineering, decomposition, multilingual/multimodal, benchmarking
- CI/CD Tools (2026): Promptfoo, DeepEval integration patterns
- Security (2026): PromptGuard 4-layer defense, Microsoft Prompt Shields, taint tracking
-
Quality Checklists - Validation checklists before deployment
- Prompt QA, JSON validation, agent workflow checks
- RAG workflow, safety & security, performance optimization
- Testing coverage, anti-patterns, quality score rubric
-
Domain-Specific Patterns - Claude 4+ optimized patterns for specialized domains
- Frontend/visual code: Creativity encouragement, design variations, micro-interactions
- Research tasks: Success criteria, verification, hypothesis tracking
- Agentic coding: No speculation rule, principled implementation, investigation patterns
- Cross-domain best practices and quality modifiers
-
核心最佳实践 - 生产级提示词的基础规则
- 系统指令设计、输出契约规范、行动指令
- 上下文处理、错误恢复、正向表述、风格匹配、风格对抗红队测试
- 反模式、Claude 4+特定优化
-
生产指南 - 部署与工程化指导
- 评估与测试(提示词CI/CD)、模型参数、少样本选择
- 安全与防护、对话记忆、上下文压缩韧性
- 答案工程、分解、多语言/多模态、基准测试
- CI/CD工具(2026年):Promptfoo、DeepEval集成模式
- 安全(2026年):PromptGuard 4层防御、Microsoft Prompt Shields、污点跟踪
-
质量检查清单 - 部署前的验证检查清单
- 提示词QA、JSON验证、Agent工作流检查
- RAG工作流、安全与合规、性能优化
- 测试覆盖、反模式、质量评分标准
-
领域特定模式 - 针对Claude 4+优化的特定领域模式
- 前端/可视化代码:创意鼓励、设计变体、微交互
- 研究任务:成功标准、验证、假设跟踪
- Agent化编码:无猜测规则、原则性实现、调查模式
- 跨领域最佳实践和质量调整
Navigation: Specialized Patterns
导航:专业模式
-
RAG Patterns - Retrieval-augmented generation workflows
- Context grounding, chunk citation, missing information handling
-
Agent and Tool Patterns - Tool use and agent orchestration
- Plan-then-act workflows, tool calling, multi-step reasoning, generate-verify-revise chains
- Multi-Agent Orchestration (2026): centralized, handoff, federated patterns; plan-and-execute (90% cost reduction)
-
Extraction Patterns - Deterministic field extraction
- Schema-based extraction, null handling, no hallucinations
-
Reasoning Patterns (Hidden CoT) - Internal reasoning without visible output
- Hidden reasoning, final answer only, classification workflows
- Extended Thinking API (Claude 4+): budget management, think tool, multishot patterns
-
Additional Patterns - Extended prompt engineering techniques
- Advanced patterns, edge cases, optimization strategies
-
RAG模式 - 检索增强生成工作流
- 上下文落地、块引用、缺失信息处理
-
Agent与工具模式 - 工具使用与Agent编排
- 先规划后执行工作流、工具调用、多步骤推理、生成-验证-修订链
- 多Agent编排(2026年):集中式、交接式、联邦式模式;规划执行模式(成本降低90%)
-
提取模式 - 确定性字段提取
- 基于Schema的提取、空值处理、无幻觉
-
推理模式(隐藏思维链) - 内部推理,无可见输出
- 隐藏推理、仅输出最终答案、分类工作流
- 扩展思考API(Claude 4+):预算管理、思考工具、少样本模式
-
其他模式 - 进阶提示词工程技术
- 高级模式、边缘案例、优化策略
Navigation: Templates
导航:模板
Templates are copy-paste ready and organized by complexity:
模板可直接复制使用,按复杂度分类:
Quick Templates
快速模板
- Quick Template - Fast, minimal prompt structure
- 快速模板 - 快速、极简的提示词结构
Standard Templates
标准模板
- Standard Template - Production-grade operational prompt
- Agent Template - Tool-using agent with planning
- RAG Template - Retrieval-augmented generation
- Chain-of-Thought Template - Hidden reasoning pattern
- JSON Extractor Template - Deterministic field extraction
- Prompt Evaluation Template - Regression tests, A/B testing, rollout gates
- 标准模板 - 生产级工程化提示词
- Agent模板 - 带规划功能的工具型Agent
- RAG模板 - 检索增强生成
- 思维链模板 - 隐藏推理模式
- JSON提取器模板 - 确定性字段提取
- 提示词评估模板 - 回归测试、A/B测试、发布门控
External Resources
外部资源
External references are listed in data/sources.json:
- Official documentation (OpenAI, Anthropic, Google)
- LLM frameworks (LangChain, LlamaIndex)
- Vector databases (Pinecone, Weaviate, FAISS)
- Evaluation tools (OpenAI Evals, HELM)
- Safety guides and standards
- RAG and retrieval resources
外部参考列于data/sources.json:
- 官方文档(OpenAI、Anthropic、Google)
- LLM框架(LangChain、LlamaIndex)
- 向量数据库(Pinecone、Weaviate、FAISS)
- 评估工具(OpenAI Evals、HELM)
- 安全指南与标准
- RAG与检索资源
Freshness Rule (2026)
时效性规则(2026年)
When asked for “latest” prompting recommendations, prefer provider docs and standards from . If web search is unavailable, state the constraint and avoid overconfident “current best” claims.
data/sources.json当被问及“最新”提示词建议时,优先参考中的厂商文档和标准。若无法进行网页搜索,需说明限制条件,避免过度自信地声称“当前最佳实践”。
data/sources.jsonRelated Skills
相关技能
This skill provides foundational prompt engineering patterns. For specialized implementations:
AI/LLM Skills:
- AI Agents Development - Production agent patterns, MCP integration, orchestration
- AI LLM Engineering - LLM application architecture and deployment
- AI LLM RAG Engineering - Advanced RAG pipelines and chunking strategies
- AI LLM Search & Retrieval - Search optimization, hybrid retrieval, reranking
- AI LLM Development - Fine-tuning, evaluation, dataset creation
Software Development Skills:
- Software Architecture Design - System design patterns
- Software Backend - Backend implementation
- Foundation API Design - API design and contracts
本技能提供基础提示词工程模式。如需专业实现,可参考:
AI/LLM技能:
- AI Agent开发 - 生产级Agent模式、MCP集成、编排
- AI LLM工程 - LLM应用架构与部署
- AI LLM RAG工程 - 进阶RAG管道与分块策略
- AI LLM搜索与检索 - 搜索优化、混合检索、重排序
- AI LLM开发 - 微调、评估、数据集创建
软件开发技能:
- 软件架构设计 - 系统设计模式
- 软件后端 - 后端实现
- 基础API设计 - API设计与契约
Usage Notes
使用说明
For Claude Code:
- Reference this skill when building prompts for agents, commands, or integrations
- Use Quick Reference table for fast pattern lookup
- Follow Decision Tree to select appropriate pattern
- Validate outputs with Quality Checklists before deployment
- Use templates as starting points, customize for specific use cases
For Codex CLI:
- Use the same patterns and templates; adapt tool-use wording to the local tool interface
- For long-horizon tasks, track progress explicitly (a step list/plan) and update it as work completes
- Run independent reads/searches in parallel when the environment supports it; keep writes/edits serialized
- AGENTS.md Integration: Place project-specific prompt guidance in AGENTS.md files at global (~/.codex/AGENTS.md), project-level (./AGENTS.md), or subdirectory scope for layered instructions
- Reasoning Effort: Use for interactive coding (default),
medium/highfor complex autonomous multi-hour tasksxhigh
针对Claude Code:
- 构建Agent、命令或集成的提示词时可参考本技能
- 使用快速参考表快速查找模式
- 遵循决策树选择合适的模式
- 部署前使用质量检查清单验证输出
- 以模板为起点,根据具体用例进行定制
针对Codex CLI:
- 使用相同的模式和模板;调整工具使用措辞以适配本地工具接口
- 对于长周期任务,需显式跟踪进度(步骤列表/计划),并在工作完成时更新
- 若环境支持,并行运行独立的读取/搜索操作;保持写入/编辑操作序列化
- AGENTS.md集成:将项目特定的提示词指导放置在全局(~/.codex/AGENTS.md)、项目级(./AGENTS.md)或子目录级的AGENTS.md文件中,实现分层指令
- 推理力度:交互式编码使用(默认),复杂的自主多小时任务使用
medium/highxhigh