improving-prompts

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Improving Prompts

提示词优化

Overview

概述

Apply documented Claude 4.5 best practices to existing prompts. Do not invent "improvements" - use the actual guidance from Anthropic.

将Anthropic官方文档中针对Claude 4.5的最佳实践应用于现有提示词。请勿自行“创造”改进方案——严格遵循Anthropic的实际指导内容。

When to Use

适用场景

Optimizing CLAUDE.md or AGENTS.md files
Improving custom command prompts
Refining skill instructions
Migrating prompts from older Claude models to 4.5

优化CLAUDE.md或AGENTS.md文件
改进自定义命令提示词
完善技能说明
将旧版Claude模型的提示词迁移至4.5版本

When NOT to Use

不适用场景

Writing new prompts from scratch (just follow best practices directly)
The prompt is working well and user hasn't identified issues

从零开始撰写新提示词（直接遵循最佳实践即可）
提示词运行正常且用户未指出任何问题

The Core Problem

核心问题

Without this skill, agents:

Invent "best practices" from general knowledge
Make structural changes without asking what's broken
Add complexity assuming more structure = better
Change things to demonstrate value rather than solve problems

如果没有此技能，智能体（Agent）会：

从通用知识中“发明”所谓的“最佳实践”
在未明确问题的情况下对提示词进行结构性修改
主观认为结构越复杂越好，无端增加复杂度
为了体现自身价值而修改内容，而非解决实际问题

Common Rationalizations (Do Not Fall For These)

常见错误借口（请勿轻信）

Rationalization	Reality
"The user said it's too vague"	"Too vague" isn't actionable. What specific behavior fails?
"I'm the expert, trust me"	Authority doesn't bypass the need for concrete issues
"Time pressure - demo tomorrow"	Pressure is when agents make the worst decisions
"The spirit of the skill is to help"	Violating the letter IS violating the spirit
"I have enough context"	If you can't name the specific failure, you don't
"Structure is always better"	Structure solves structure problems, not all problems
"This is obviously an improvement"	Obvious to you ≠ solving the user's actual problem

错误借口	实际情况
“用户说这个提示词太模糊了”	“太模糊”不具备可操作性。具体是哪部分行为不符合预期？
“我是专家，相信我”	权威不能替代对具体问题的明确界定
“时间紧迫——明天要演示”	压力往往是智能体做出最差决策的时候
“这个技能的初衷就是提供帮助”	违反规则本身就是违背初衷
“我已经掌握足够上下文了”	如果你无法明确指出具体的问题，说明你并没有足够上下文
“结构化总是更好的”	结构化只能解决结构性问题，而非所有问题
“这显然是一个改进”	你觉得明显≠解决用户的实际问题

Required Process

必选流程

Step 1: Understand Before Changing

步骤1：先理解再修改

Before ANY modifications:

Ask what specific behaviors are underperforming
Ask what the prompt should achieve that it currently doesn't
If user says "just improve it generally" - ask for at least one concrete issue

What counts as a "concrete issue":

"Claude ignores my instruction to be concise" ✓
"The examples I provide don't match the output format" ✓
"Claude suggests changes but doesn't implement them" ✓

What does NOT count:

"It's too vague" ✗ (vague about what?)
"It doesn't follow best practices" ✗ (which practice? what fails?)
"It's inconsistent" ✗ (inconsistent how? show examples)

Do NOT proceed with generic "improvements" based on assumptions.

在进行任何修改之前：

询问用户具体哪些表现未达预期
询问用户期望提示词实现但当前未实现的目标
如果用户说“只是整体优化一下”——请至少让用户指出一个具体问题

什么是“具体问题”：

“Claude无视我要求简洁的指令” ✓
“我提供的示例与输出格式不匹配” ✓
“Claude只提出修改建议但不执行” ✓

什么不属于“具体问题”：

“它太模糊了” ✗（具体哪里模糊？）
“它不符合最佳实践” ✗（哪项实践？具体哪里不符合？）
“它不一致” ✗（怎么不一致？请举例说明）

请勿基于假设进行笼统的“改进”。

Step 2: Reference Actual Best Practices

步骤2：参考官方最佳实践

See

references/claude-4.5-best-practices.md

for the complete reference. Key principles:

Be explicit with instructions:

Claude 4.5 follows instructions precisely - vague requests get literal interpretations
If you want "above and beyond" behavior, explicitly request it
Example: "Create a dashboard" → "Create a dashboard. Include relevant features and interactions. Go beyond basics."

Add context/motivation:

Explain WHY a rule exists, not just WHAT the rule is
Claude generalizes from explanations
Example: "NEVER use ellipses" → "Never use ellipses because the text-to-speech engine cannot pronounce them"

Be vigilant with examples:

Examples are followed precisely - ensure they demonstrate desired behavior
One excellent example beats many mediocre ones

Avoid "think" without extended thinking:

When extended thinking is disabled, "think" triggers unwanted behavior
Use alternatives: "consider," "evaluate," "assess," "determine"

Control verbosity explicitly:

Claude 4.5 defaults to efficiency/conciseness
If you want summaries or explanations, request them explicitly

Tool usage patterns:

"Can you suggest changes" → Claude suggests but doesn't implement
"Make these changes" → Claude implements
Be explicit about whether to act or advise

完整参考内容请查看

references/claude-4.5-best-practices.md

。核心原则如下：

指令要明确：

Claude 4.5会精准遵循指令——模糊的请求会被字面解读
如果你想要“超出基础要求”的表现，请明确提出
示例：“创建一个仪表盘” → “创建一个仪表盘，包含相关功能和交互，超出基础要求”

添加背景/动机：

解释规则存在的原因，而不仅仅是规则内容
Claude会从解释中进行泛化学习
示例：“绝对不要使用省略号” → “绝对不要使用省略号，因为文本转语音引擎无法识别它们”

谨慎使用示例：

Claude会严格遵循示例——确保示例能展示期望的行为
一个优秀的示例胜过多个平庸的示例

避免无扩展思考的“think”指令：

当扩展思考功能被禁用时，“think”会触发意外行为
使用替代词汇：“consider”“evaluate”“assess”“determine”

明确控制冗长程度：

Claude 4.5默认以高效/简洁为原则
如果你需要总结或详细解释，请明确提出

工具使用模式：

“Can you suggest changes”（你能提出修改建议吗？）→ Claude只会提出建议，不会执行
“Make these changes”（执行这些修改）→ Claude会执行修改
明确说明是要提供建议还是直接执行

Step 3: Preserve What Works

步骤3：保留有效的内容

Do NOT restructure sections that aren't problematic
Do NOT add complexity unless solving a stated problem
Do NOT change tone/voice unless specifically requested
Keep the user's examples if they demonstrate the right behavior

请勿对没有问题的章节进行结构调整
除非是为了解决明确的问题，否则不要增加复杂度
除非用户明确要求，否则不要改变语气/风格
如果用户提供的示例能展示正确行为，请保留

Step 4: Propose Changes with Rationale

步骤4：附带理由提出修改方案

For each change, state:

What specific best practice it applies
What problem it solves
Show before/after

Do NOT make changes without connecting them to documented guidance.

对于每一项修改，需要说明：

应用了哪项具体的最佳实践
解决了什么问题
展示修改前后的对比

请勿在未关联官方文档指导内容的情况下进行修改。

Red Flags - You're About to Fail

危险信号——你即将犯错

"Based on general best practices..." → STOP. Use documented practices.
"Structure is always better..." → STOP. Ask if structure is the problem.
"I'll assume the user wants..." → STOP. Ask.
Making 10+ changes to a short prompt → STOP. What specific problem are you solving?
"This is how I would write it..." → STOP. You're not the user.

“基于通用最佳实践……” → 停止。请使用官方文档中的实践。
“结构化总是更好……” → 停止。先询问是否是结构问题。
“我假设用户想要……” → 停止。去询问用户。
对简短的提示词进行10+项修改 → 停止。你到底在解决什么具体问题？
“我会这么写……” → 停止。你不是用户。

Quick Reference: Claude 4.5 Improvements

快速参考：Claude 4.5提示词改进方案

Issue	Fix
Claude takes things too literally	Add "Go beyond basics" or explicit scope
Claude doesn't explain reasoning	Add "Explain your reasoning" or "Think through this step by step"
Claude is too verbose	Add "Be concise" or "Respond in X sentences"
Claude is too terse	Add "Provide detailed explanations"
Claude suggests but doesn't act	Change "Can you..." to imperative "Do X"
Instruction isn't followed	Add context for WHY the instruction matters
Examples not matching output	Ensure examples show exact desired format

问题	解决方案
Claude过于字面化解读指令	添加“超出基础要求”或明确的范围说明
Claude不解释推理过程	添加“Explain your reasoning”（解释你的推理过程）或“Think through this step by step”（逐步思考）
Claude过于冗长	添加“Be concise”（保持简洁）或“Respond in X sentences”（用X句话回复）
Claude过于简洁	添加“Provide detailed explanations”（提供详细解释）
Claude只提建议不执行	将“Can you...”改为祈使句“Do X”（执行X操作）
指令未被遵循	添加指令背后的原因背景
示例与输出不匹配	确保示例展示了完全符合要求的格式

Common Mistakes

常见错误

Overengineering: Adding categories, numbered lists, XML structure to simple prompts that were working fine.

Changing voice: The user's CLAUDE.md reflects their personality. Don't make it sound like documentation.

Assuming problems: Making changes without knowing what's actually broken.

Inventing practices: Claiming something is a "best practice" without reference to actual guidance.

过度设计： 为原本运行良好的简单提示词添加分类、编号列表、XML结构等。

改变风格： 用户的CLAUDE.md体现了他们的个人风格，不要让它听起来像官方文档。

假设问题： 在不知道实际问题的情况下进行修改。

自创实践： 在未参考官方指导的情况下，声称某内容是“最佳实践”。