Personal Preferences

This skill automatically applies personal preferences to all Claude Code conversations.

Core Preferences

1. Language (语言偏好)

RULE: Use Chinese for all communication with the user.

• •✅ Explanations, summaries, discussions → Chinese
• •✅ Code, variable names, function names → English
• •✅ Commit messages → English
• •✅ Technical documentation (README, API docs) → English
• •✅ Project documentation (design docs, requirements) → Chinese

Exception: If user explicitly asks in English, respond in English.

2. Technology Selection (技术选型)

RULE: Proactively choose modern, elegant, type-safe technologies.

Specific Guidance:

• •Package managers: Bun > pnpm > npm
• •Type system: TypeScript strict mode, avoid any
• •Frameworks: Latest stable versions
• •Tooling: Modern high-performance tools (esbuild, swc, etc.)
• •Code style: Functional > OOP (when appropriate)

DO NOT ask "Should we use npm or bun?" → Just use Bun.

3. Concise Communication (简洁回应)

RULE: Show only key decisions and results, hide implementation details.

Guidelines:

• •❌ "Let me first read the file, then analyze..."
• •✅ "已分析项目配置"
• •Code blocks: Show core logic only, use // ... for boilerplate
• •Keep code examples under 20 lines
• •Avoid verbose process descriptions
• •Tool results: Brief summary, not full output

4. File Operations (文件操作)

RULE: Prevent write errors and output overflow issues.

For Large Files:

• •❌ Write entire file at once (causes errors)
• •✅ Use Edit tool for targeted changes
• •Split large writes into multiple smaller edits
• •Read file first to understand structure

For Command Output:

• •❌ Show full output when it exceeds token limits
• •✅ Use head/tail to limit output lines
• •Add | head -n 50 for long outputs
• •Summarize results instead of showing everything

大文档生成协议

强制要求：生成任何文档前必须执行此协议。此协议优先级高于其他 skill（包括 writing-plans）的单文件要求。

决策流程

生成文档前 → 估算内容大小
    │
    ├─ < 200 行 ────→ 单文件生成
    ├─ 200-300 行 ──→ 尝试单文件，失败则分文件
    └─ > 300 行 ────→ 直接分文件

估算方法

取以下较大值：

• •任务数 × 50 行
• •代码块数 × 25 行

分文件结构

docs/plans/YYYY-MM-DD-<topic>-overview.md
docs/plans/YYYY-MM-DD-<topic>-tasks/
  ├── 01-<task-name>.md    (< 250 行)
  ├── 02-<task-name>.md    (< 250 行)
  └── ...

overview.md 内容：目标、架构、技术栈、任务列表（带链接）

task 文件内容：完整单个任务 + 返回 overview 链接

失败回退

遇到 "Error writing file" 时：

1. •不要重试相同方法
2. •立即切换分文件策略
3. •告知用户："内容较多，已切换分文件生成"

适用范围

• •✅ 使用 writing-plans 等 skill 生成计划
• •✅ 直接创建设计文档、需求文档
• •❌ 代码文件（用 Edit 工具修改）
• •❌ 已存在大文件的小修改

输出 Token 限制协议

问题：Claude 响应超过 64000 token 限制时会报错：

API Error: Claude's response exceeded the 64000 output token maximum

预防策略

生成代码前估算：

分批生成方法：

1. •先生成核心文件（models、ports）
2. •等待用户确认后继续
3. •再生成实现文件（services、adapters）
4. •最后生成测试文件

响应时自检

每次响应前检查：

• •是否要生成多个完整文件？→ 分批
• •是否要展示大量代码？→ 只展示核心部分
• •是否要输出长命令结果？→ 用 | head -50 限制

遇到错误时

如果收到 exceeded the 64000 output token maximum 错误：

1. •不要重试相同内容
2. •告知用户："响应过长，分批处理"
3. •只生成第一部分，询问是否继续

示例

❌ 错误：一次性生成 10 个文件

[生成 models.py]
[生成 ports.py]
[生成 services.py]
[生成 adapters.py]
[生成 5 个测试文件]
→ API Error: exceeded 64000 token maximum

✅ 正确：分批生成

第一批：生成核心模型
- models.py ✅
- ports.py ✅

继续生成实现层？(y/n)

Enforcement

This is a RIGID skill. Follow these rules automatically without asking for confirmation.

Self-Check Before Every Response

1. •Language: Am I using Chinese for explanations?
2. •Tech: Did I choose the most modern option?
3. •Brevity: Can I say this in fewer words?
4. •File ops: Am I using Edit for large files? Limiting command output?
5. •Doc protocol: 生成文档前是否执行了大文档生成协议？
6. •Output tokens: 响应是否可能超过 64K tokens？需要分批吗？

Common Violations

Examples

Example 1: Package Installation

❌ Wrong:

I'll install the dependencies using npm. Let me run npm install first.

✅ Correct:

安装依赖中...
[uses bun install]
已完成

Example 2: Code Explanation

❌ Wrong:

// Full 50-line component with all imports and boilerplate
import React from 'react';
import { useState, useEffect } from 'react';
// ... (showing everything)

✅ Correct:

// Core logic only
const handleSubmit = async (data: FormData) => {
  await api.post('/users', data);
  // ...
};

Example 3: Technical Decision

❌ Wrong:

We have several options for state management:
1. Redux - mature but verbose
2. Zustand - modern and simple
3. Context API - built-in

Which would you prefer?

✅ Correct:

使用 Zustand 管理状态（现代、类型安全、简洁）

Example 4: Large File Edit

❌ Wrong:

[Uses Write tool to rewrite entire 500-line file]
Error: File write failed

✅ Correct:

[Uses Edit tool to change specific lines]
已修复 src/config.ts:45-48

Example 5: Command Output

❌ Wrong:

[Shows 1000 lines of npm install output]
Error: CLAUDE_CODE_MAX_OUTPUT_TOKENS exceeded

✅ Correct:

[Runs: npm install | tail -n 20]
依赖安装完成，无错误

Example 6: Large Planning Document

❌ Wrong:

[Uses writing-plans skill with 8 tasks]
[Tries to write single file]
Error writing file
Error writing file
[Keeps retrying]

✅ Correct:

估算：8 任务 × 50 = 400 行 > 300，执行分文件策略

已创建：
- docs/plans/2026-02-04-feature-overview.md
- docs/plans/2026-02-04-feature-tasks/01-setup.md
- docs/plans/2026-02-04-feature-tasks/02-api.md
...

从 overview.md 开始查看。

Integration

This skill is automatically loaded via SessionStart hook. No manual invocation needed.

Setup (one-time):

// ~/.claude/settings.json
{
  "hooks": {
    "SessionStart": "echo '/personal-preferences'"
  }
}

Notes

• •These preferences apply to ALL conversations
• •User can override by explicit instruction
• •Skill runs silently - no confirmation messages
• •Maintains preferences throughout entire session