Back to Details

context-compressor

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Context Compressor Skill

Context Compressor Skill

<identity> Context Compressor Skill - Techniques for reducing token usage while preserving decision-critical information. Helps agents work efficiently within context limits. </identity> <capabilities> - Compressing conversation history - Summarizing code and documentation - Extracting key decisions and context - Creating efficient memory snapshots - Reducing redundancy in context </capabilities> <instructions> <execution_process>
<identity> Context Compressor Skill - 用于在保留决策关键信息的同时降低Token使用量的技巧，帮助Agent在上下文限制内高效工作。 </identity> <capabilities> - 压缩对话历史 - 总结代码与文档 - 提取关键决策与上下文信息 - 创建高效的内存快照 - 减少上下文中的冗余内容 </capabilities> <instructions> <execution_process>

Step 1: Identify Compressible Content

步骤1：识别可压缩内容

Content types that can be compressed:
TypeCompression Strategy
CodeKeep signatures, summarize implementations
ConversationsExtract decisions, drop small talk
DocumentationKeep headings and key points
ErrorsKeep message and location, drop stack frames
LogsKeep patterns, drop repetitions
可压缩的内容类型：
类型压缩策略
代码保留签名，总结实现逻辑
对话内容提取决策信息，剔除闲聊内容
文档保留标题与核心要点
错误信息保留错误消息与位置，剔除调用栈帧
日志保留模式规律，剔除重复内容

Step 2: Apply Compression Techniques

步骤2：应用压缩技巧

Technique 1: Decision Extraction
Before:
User: Should we use Redis or Memcached?
Assistant: Let me analyze both options...
[500 words of analysis]
Recommendation: Redis for pub/sub support.
User: Ok let's use Redis.
After:
Decision: Use Redis (chosen for pub/sub support)
Technique 2: Code Summarization
Before:
javascript
// 100 lines of UserService implementation
After:
UserService: CRUD operations for users
- Methods: create, read, update, delete, findByEmail
- Dependencies: db, validator, logger
- Location: src/services/user.js
Technique 3: Error Compression
Before:
Error: Cannot read property 'id' of undefined
    at UserController.getUser (src/controllers/user.js:45:23)
    at Layer.handle [as handle_request] (node_modules/express/lib/router/layer.js:95:5)
    ... 20 more stack frames
After:
Error: Cannot read 'id' of undefined @ src/controllers/user.js:45
Cause: User object is null when accessing .id
技巧1：决策提取
压缩前：
User: Should we use Redis or Memcached?
Assistant: Let me analyze both options...
[500 words of analysis]
Recommendation: Redis for pub/sub support.
User: Ok let's use Redis.
压缩后：
Decision: Use Redis (chosen for pub/sub support)
技巧2：代码总结
压缩前：
javascript
// 100 lines of UserService implementation
压缩后：
UserService: 用户相关CRUD操作
- 方法：create、read、update、delete、findByEmail
- 依赖：db、validator、logger
- 位置：src/services/user.js
技巧3：错误信息压缩
压缩前：
Error: Cannot read property 'id' of undefined
    at UserController.getUser (src/controllers/user.js:45:23)
    at Layer.handle [as handle_request] (node_modules/express/lib/router/layer.js:95:5)
    ... 20 more stack frames
压缩后：
Error: Cannot read 'id' of undefined @ src/controllers/user.js:45
原因：访问.id时User对象为null

Step 3: Structure Compressed Output

步骤3：结构化压缩输出

Use consistent formats:
markdown
undefined
使用统一格式：
markdown
undefined

Session Summary

会话总结

Decisions Made

已做出的决策

  • [D1] Use Redis for caching
  • [D2] JWT for authentication
  • [D1] 使用Redis作为缓存
  • [D2] 使用JWT进行身份验证

Files Modified

修改的文件

  • src/auth/jwt.js (new)
  • src/config/redis.js (updated)
  • src/auth/jwt.js（新增）
  • src/config/redis.js（更新）

Open Items

待办事项

  • Add rate limiting
  • Write tests for JWT
undefined
  • 添加限流机制
  • 为JWT编写测试用例
undefined

Step 4: Validate Compression

步骤4：验证压缩效果

Ensure critical info preserved:
  • All decisions captured
  • Key file locations retained
  • Error causes documented
  • Next steps clear
</execution_process>
<best_practices>
  1. Preserve Decisions: Never lose decision rationale
  2. Keep Locations: File paths and line numbers are critical
  3. Summarize, Don't Delete: Transform verbose content
  4. Use References: Point to files instead of including content
  5. Test Recovery: Can you continue work from compressed context?
</best_practices> </instructions>
<examples> <usage_example> **Compression Request**: 
Compress the last 10 messages for context handoff
Example Response:
markdown
undefined
确保关键信息已保留：
  • 所有决策已捕获
  • 关键文件位置已保留
  • 错误原因已记录
  • 后续步骤清晰明确
</execution_process>
<best_practices>
  1. 保留决策信息：绝不能丢失决策的依据
  2. 保留位置信息：文件路径与行号至关重要
  3. 总结而非删除：将冗长内容进行转化
  4. 使用引用：指向文件而非直接包含内容
  5. 测试可恢复性：能否基于压缩后的上下文继续开展工作？
</best_practices> </instructions>
<examples> <usage_example> **压缩请求**： 
Compress the last 10 messages for context handoff
示例响应：
markdown
undefined

Context Summary (compressed from 10 messages)

上下文总结（由10条消息压缩而来）

Task

任务

Implementing user authentication with JWT
基于JWT实现用户身份验证

Progress

进度

  1. Created JWT utility (src/auth/jwt.js)
  2. Added login endpoint (src/routes/auth.js)
  3. Tests failing - mock issue
  1. 已创建JWT工具类（src/auth/jwt.js）
  2. 已添加登录接口（src/routes/auth.js）
  3. 测试失败 - 模拟数据存在问题

Key Decisions

关键决策

  • Using RS256 for JWT signing (security over performance)
  • 1-hour token expiry with refresh tokens
  • 使用RS256进行JWT签名（优先考虑安全性而非性能）
  • Token有效期为1小时，支持刷新Token

Current Issue

当前问题

Jest mock for jwt.sign not working Location: src/auth/tests/jwt.test.js:23
Jest对jwt.sign的模拟不生效 位置：src/auth/tests/jwt.test.js:23

Next Steps

后续步骤

  1. Fix JWT mock
  2. Add refresh token endpoint

</usage_example>
</examples>
  1. 修复JWT模拟问题
  2. 添加刷新Token接口

</usage_example>
</examples>

Rules

规则

  • Never lose decision rationale
  • Always include file locations
  • Test that work can continue from compressed context
  • 绝不能丢失决策的依据
  • 始终包含文件位置信息
  • 验证能否基于压缩后的上下文继续开展工作

Workflow Integration

工作流集成

This skill supports multi-agent orchestration by enabling efficient context management:
Router Decision: 
.claude/workflows/core/router-decision.md
  • Router spawns agents that use this skill for context-efficient handoffs
  • Used in long-running sessions to maintain continuity
Artifact Lifecycle: 
.claude/workflows/core/skill-lifecycle.md
  • Compression patterns evolve with framework changes
  • Session summaries feed into memory protocol
Related Workflows:
  • session-handoff
    skill for complete handoff protocol
  • swarm-coordination
    skill for multi-agent context sharing
该Skill通过高效的上下文管理支持多Agent编排：
路由决策：
.claude/workflows/core/router-decision.md
  • 路由模块会生成使用该Skill的Agent，以实现高效的上下文交接
  • 用于长时运行会话，维持工作连续性
工件生命周期：
.claude/workflows/core/skill-lifecycle.md
  • 压缩模式会随框架更新而演进
  • 会话总结会同步到内存协议中
相关工作流：
  • session-handoff
    Skill：用于完整的上下文交接协议
  • swarm-coordination
    Skill：用于多Agent上下文共享

Memory Protocol (MANDATORY)

内存协议（强制要求）

Before starting:
bash
cat .claude/context/memory/learnings.md
After completing:
  • New pattern -> 
    .claude/context/memory/learnings.md
  • Issue found -> 
    .claude/context/memory/issues.md
  • Decision made -> 
    .claude/context/memory/decisions.md
ASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.
开始前：
bash
cat .claude/context/memory/learnings.md
完成后：
  • 新的模式 -> 
    .claude/context/memory/learnings.md
  • 发现的问题 -> 
    .claude/context/memory/issues.md
  • 做出的决策 -> 
    .claude/context/memory/decisions.md
假设可能出现中断：你的上下文可能会重置。如果信息未存入内存，则视为未发生。