context-engineering-kit

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Context Engineering Kit

上下文工程工具包

A collection of advanced context engineering techniques and patterns designed to improve AI agent results while minimizing token consumption.
这是一套高级上下文工程技术与模式集合,旨在提升AI Agent的输出效果,同时最大限度减少Token消耗。

When to Use This Skill

何时使用该技能

  • Improving AI output quality systematically
  • Reducing token usage in complex tasks
  • Implementing structured reasoning patterns
  • Multi-agent code review workflows
  • Spec-driven development processes
  • When standard prompting isn't enough
  • 系统性提升AI输出质量
  • 减少复杂任务中的Token使用量
  • 实现结构化推理模式
  • 多Agent代码审查工作流
  • 规范驱动的开发流程
  • 当标准提示词无法满足需求时

Core Techniques

核心技术

1. Reflexion Pattern

1. Reflexion模式

Feedback loops that improve output by 8-21% across tasks.
How it works:
1. Generate initial response
2. Self-evaluate against criteria
3. Identify improvement areas
4. Generate refined response
5. Repeat until quality threshold met
Use when:
  • Writing complex code
  • Creating documentation
  • Solving multi-step problems
通过反馈循环可将各类任务的输出效果提升8-21%。
工作原理:
1. 生成初始响应
2. 根据标准进行自我评估
3. 识别可改进区域
4. 生成优化后的响应
5. 重复直至达到质量阈值
适用场景:
  • 编写复杂代码
  • 创建文档
  • 解决多步骤问题

2. Spec-Driven Development

2. 规范驱动开发

Based on GitHub Spec Kit and OpenSpec frameworks.
Process:
markdown
undefined
基于GitHub Spec Kit和OpenSpec框架。
流程:
markdown
undefined

Specification Document

规范文档

Requirements

需求

[Clear, testable requirements]
[清晰、可测试的需求]

Acceptance Criteria

验收标准

[Specific success conditions]
[具体的成功条件]

Constraints

约束条件

[Limitations and boundaries]
[限制与边界]

Examples

示例

[Input/output pairs]

**Benefits:**
- Clearer requirements
- Testable outputs
- Reduced ambiguity
[输入/输出配对]

**优势:**
- 需求更清晰
- 输出可测试
- 减少歧义

3. Subagent-Driven Development

3. 子Agent驱动开发

Competitive generation with quality gates.
Architecture:
┌─────────────────────────┐
│   Orchestrator Agent    │
├─────────────────────────┤
│  ┌─────┐  ┌─────┐      │
│  │Gen 1│  │Gen 2│ ...  │
│  └─────┘  └─────┘      │
├─────────────────────────┤
│    Quality Gate Agent   │
└─────────────────────────┘
Flow:
  1. Multiple agents generate solutions
  2. Quality agent evaluates each
  3. Best solution selected/merged
  4. Iterative refinement
带质量关卡的竞争性生成机制。
架构:
┌─────────────────────────┐
│   Orchestrator Agent    │
├─────────────────────────┤
│  ┌─────┐  ┌─────┐      │
│  │Gen 1│  │Gen 2│ ...  │
│  └─────┘  └─────┘      │
├─────────────────────────┤
│    Quality Gate Agent   │
└─────────────────────────┘
流程:
  1. 多个Agent生成解决方案
  2. 质量Agent评估每个方案
  3. 选择/合并最佳解决方案
  4. 迭代优化

4. First Principles Framework (FPF)

4. 第一性原理框架(FPF)

Hypothesis-driven decision making.
Structure:
Observation → Hypothesis → Test → Conclusion
Application:
  • Debugging complex issues
  • Architecture decisions
  • Technology selection
基于假设的决策制定。
结构:
观察 → 假设 → 测试 → 结论
应用场景:
  • 调试复杂问题
  • 架构决策
  • 技术选型

5. Kaizen (Continuous Improvement)

5. Kaizen(持续改进)

Systematic iterative enhancement.
Cycle:
Plan → Do → Check → Act → Repeat
系统性的迭代增强机制。
循环:
计划 → 执行 → 检查 → 改进 → 重复

Plugin Categories

插件分类

Reasoning Enhancement

推理增强

  • Reflexion: Self-evaluation loops
  • Chain of Thought: Step-by-step reasoning
  • Tree of Thoughts: Branching exploration
  • Reflexion: 自我评估循环
  • Chain of Thought: 逐步推理
  • Tree of Thoughts: 分支探索

Code Quality

代码质量

  • Multi-Agent Review: Parallel code analysis
  • Security Audit: Vulnerability detection
  • Performance Analysis: Optimization suggestions
  • Multi-Agent Review: 并行代码分析
  • Security Audit: 漏洞检测
  • Performance Analysis: 优化建议

Development Process

开发流程

  • Spec-Driven: Requirements-first approach
  • TDD Support: Test-first workflows
  • Documentation: Auto-generated docs
  • Spec-Driven: 需求优先方法
  • TDD Support: 测试优先工作流
  • Documentation: 自动生成文档

Meta-Skills

元技能

  • Plugin Development: Create new plugins
  • Workflow Composition: Combine techniques
  • Performance Tuning: Optimize patterns
  • Plugin Development: 创建新插件
  • Workflow Composition: 组合技术
  • Performance Tuning: 优化模式

How to Use

使用方法

Basic: Reflexion Loop

基础用法:Reflexion循环

Review my code with reflexion:

[paste code]

Requirements:
- Error handling
- Performance
- Readability
使用Reflexion审查我的代码:

[粘贴代码]

需求:
- 错误处理
- 性能
- 可读性

Spec-Driven Task

规范驱动任务

Create a spec for: User authentication system

Then implement following the spec.
为用户认证系统创建一份规范文档。

然后按照规范实现。

Multi-Agent Review

多Agent审查

Review this PR with multiple perspectives:
- Security focus
- Performance focus
- Maintainability focus

[paste code or PR link]
从多个角度审查此PR:
- 安全聚焦
- 性能聚焦
- 可维护性聚焦

[粘贴代码或PR链接]

Token Efficiency Tips

Token效率技巧

1. Structured Prompts

1. 结构化提示词

markdown
undefined
markdown
undefined

Context

上下文

[Brief, relevant context only]
[仅包含简短、相关的上下文]

Task

任务

[Clear, specific task]
[清晰、具体的任务]

Output Format

输出格式

[Expected structure]
undefined
[预期的结构]
undefined

2. Progressive Disclosure

2. 渐进式披露

  • Start with essential info
  • Add details only when needed
  • Remove redundant context
  • 从必要信息开始
  • 仅在需要时添加细节
  • 移除冗余上下文

3. Pattern Libraries

3. 模式库

  • Reuse proven patterns
  • Reference by name
  • Avoid repeated explanations
  • 复用已验证的模式
  • 通过名称引用
  • 避免重复解释

Example: Complex Code Review

示例:复杂代码审查

Traditional approach (~2000 tokens):
Review this code for bugs, security issues, performance problems...
Context-engineered approach (~800 tokens):
markdown
undefined
传统方法(约2000 Token):
审查此代码中的Bug、安全问题、性能问题...
上下文工程优化方法(约800 Token):
markdown
undefined

Review: auth.py

审查:auth.py

Focus Areas

聚焦领域

  1. Security (OWASP Top 10)
  2. Error handling
  3. SQL injection
  1. 安全(OWASP Top 10)
  2. 错误处理
  3. SQL注入

Output

输出

  • Issues: severity + line number
  • Fixes: specific code suggestions
[code]

**Result:** Same quality, 60% fewer tokens.
  • 问题:严重程度 + 行号
  • 修复:具体代码建议
[代码]

**结果:** 质量相同,Token用量减少60%。

Best Practices

最佳实践

  1. Start Simple: Add complexity only when needed
  2. Measure Impact: Track quality improvements
  3. Iterate: Refine patterns based on results
  4. Document: Keep notes on what works
  5. Share: Contribute successful patterns
  1. 从简开始: 仅在需要时增加复杂度
  2. 衡量影响: 跟踪质量提升情况
  3. 迭代优化: 根据结果改进模式
  4. 文档记录: 记录有效的方法
  5. 分享交流: 贡献成功的模式

Integration

集成兼容性

Works with:
  • Claude Code
  • Cursor
  • VS Code + Continue
  • Any LLM-based tool
可与以下工具配合使用:
  • Claude Code
  • Cursor
  • VS Code + Continue
  • 任何基于LLM的工具

Creating Custom Patterns

创建自定义模式

markdown
undefined
markdown
undefined

Pattern: [Name]

模式:[名称]

When to Use

何时使用

[Trigger conditions]
[触发条件]

Process

流程

[Step-by-step]
[分步说明]

Example

示例

[Concrete example]
[具体示例]

Metrics

指标

[How to measure success]
undefined
[如何衡量成功]
undefined