context-optimizer
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseContext Optimizer
上下文优化器
Manage your context window and token budget effectively.
有效管理你的上下文窗口和 Token 预算。
Quick Diagnosis
快速诊断
- Run to check current usage
/context - If > 70% → compact now before it degrades
- If > 90% → you're in the "dumb zone", compact immediately
- 运行 检查当前使用量
/context - 如果 > 70% → 在情况恶化前立即压缩
- 如果 > 90% → 你已进入“低效区”,立即压缩
Optimization Strategies
优化策略
Immediate
立即执行
| Action | Saves | When |
|---|---|---|
| 30-50% context | At task boundaries |
| Disable unused MCPs | ~5% per MCP | When switching domains |
| Use subagents for exploration | Keeps main context clean | Heavy search/read tasks |
Fresh session via | 100% reset | When starting unrelated work |
| 操作 | 可节省 | 适用场景 |
|---|---|---|
| 30-50% 上下文 | 任务边界处 |
| 禁用未使用的 MCP | 每个 MCP 约 5% | 切换领域时 |
| 使用 subagent 进行探索 | 保持主上下文整洁 | 搜索或读取任务量较大时 |
通过 | 100% 重置 | 开始无关工作时 |
Configuration
配置
Set proactive auto-compaction:
json
{
"env": {
"CLAUDE_AUTOCOMPACT_PCT_OVERRIDE": "50"
}
}设置主动自动压缩:
json
{
"env": {
"CLAUDE_AUTOCOMPACT_PCT_OVERRIDE": "50"
}
}MCP Audit
MCP 审计
Keep <10 MCPs enabled, <80 tools total. Each MCP adds overhead to every request.
bash
/mcp # List active servers启用的 MCP 应少于 10 个,工具总数应少于 80 个。每个 MCP 都会为每次请求增加开销。
bash
/mcp # List active serversDisable what you're not using
Disable what you're not using
undefinedundefinedPrompt Engineering for Token Efficiency
提升 Token 效率的 Prompt 工程
- Scope your prompts: "In src/auth/, fix the login bug"
- Provide constraints: "Don't modify the middleware"
- Give acceptance criteria: "Should return 429 after 5 attempts"
- Avoid vague prompts: "Fix the code" (forces Claude to read everything)
- 限定 Prompt 范围:“在 src/auth/ 中,修复登录缺陷”
- 提供约束:“不要修改 middleware”
- 给出验收标准:“尝试 5 次后应返回 429”
- 避免模糊 Prompt:“修复代码”(这会迫使 Claude 读取全部内容)
Subagent Delegation
Subagent 委派
Heavy operations that generate lots of output should go to subagents:
- Test suite output → subagent
- Large file exploration → subagent
- Documentation generation → subagent
- Log analysis → subagent
The main session stays clean while subagents handle the volume.
应将产生大量输出的高开销操作交给 subagent:
- 测试套件输出 → subagent
- 大型文件探索 → subagent
- 文档生成 → subagent
- 日志分析 → subagent
主会话保持整洁,由 subagent 处理大量输出。
Context Budget Planning
上下文预算规划
| Phase | Target Usage | Action If Over |
|---|---|---|
| Planning | < 20% | Keep plans concise |
| Implementation | < 60% | Compact between files |
| Testing | < 80% | Delegate to subagent |
| Review | < 90% | Start fresh session |
| 阶段 | 目标使用率 | 超出时的操作 |
|---|---|---|
| 规划 | < 20% | 保持方案简洁 |
| 实施 | < 60% | 在文件之间压缩上下文 |
| 测试 | < 80% | 委派给 subagent |
| 审查 | < 90% | 开启全新会话 |
Token Efficiency
Token 效率
Output Reduction (40-60% savings)
输出精简(节省 40-60%)
- No sycophantic openers ("Sure!", "Great question!")
- No closing fluff ("Let me know if you need anything!")
- No prompt restatement before answering
- Code first, explanation only if non-obvious
- Structured output (tables, bullets) over prose
- ASCII only: -- not em dashes, straight quotes not smart quotes
- 不使用谄媚式开场白(“Sure!”、“Great question!”)
- 不使用收尾废话(“Let me know if you need anything!”)
- 回答前不要复述 Prompt
- 代码优先,仅在含义不直观时进行解释
- 使用结构化输出(表格、项目符号)而非长篇文本
- 仅使用 ASCII:使用 --,不要使用长破折号;使用直引号,不要使用智能引号
Behavioral Efficiency
行为效率
- One-pass coding: complete solution, test once, stop if green
- Read before write: never modify unread files
- No re-reads: don't re-read unchanged files
- Tool-call budgets: 20 (quick fix) to 80 (large feature)
- Never iterate more than twice on the same failure
- 一次完成编码:提供完整解决方案,测试一次,通过后停止
- 先读后写:绝不修改未读取的文件
- 不要重复读取:不要重新读取未变更的文件
- 工具调用预算:快速修复为 20,大型功能为 80
- 对同一失败绝不迭代超过两次
Task Profiles
任务类型
Switch response style based on context:
- Coding: code first, minimal explanation, simplest solution
- Agent/Pipeline: structured output only, no prose, parseable
- Analysis: finding first, tables over paragraphs, sourced numbers
根据上下文切换响应风格:
- 编码:代码优先,解释最少,采用最简单的解决方案
- Agent/Pipeline:仅输出结构化内容,不要长篇说明,确保可解析
- 分析:先给结论,表格优于段落,数字需有来源
CLAUDE.md Optimization
CLAUDE.md 优化
- Root CLAUDE.md: < 60 lines ideal, < 150 max
- Move package-specific info to package-level CLAUDE.md
- Move personal preferences to CLAUDE.local.md
- Remove obvious or rapidly-changing information
- 根目录 CLAUDE.md:理想情况下少于 60 行,最多少于 150 行
- 将包特定信息移到包级 CLAUDE.md
- 将个人偏好移到 CLAUDE.local.md
- 移除显而易见或快速变化的信息
When Context Is Degraded
上下文退化时
Signs:
- Claude repeats itself or forgets earlier context
- Responses become generic or lose project-specific knowledge
- Tool calls start failing for reasons that worked earlier
Fix:
- Manual
/compact - If still bad: new session with
/resume - For recurring issues: reduce CLAUDE.md size, disable MCPs
迹象:
- Claude 重复自身内容或忘记之前的上下文
- 响应变得泛化或丢失项目特定知识
- 此前正常的工具调用开始失败
修复方法:
- 手动执行
/compact - 如果仍然不佳:通过 开启新会话
/resume - 如果问题反复出现:缩减 CLAUDE.md 大小,禁用 MCP