Claude Code 内存优化器：4步减少30-50%令牌，提升加载速度与API成本效益

memory-optimize by kochetkov-ma/claude-brewcode

1,100 周安装量

16 GitHub Stars

GitHub

安装命令

npx skills add https://github.com/kochetkov-ma/claude-brewcode --skill memory-optimize

AI/机器学习开发生产力

🇨🇳中文介绍

插件：kochetkov-ma/claude-brewcode

内存优化器

通过 4 个交互步骤 优化 Claude Code 的 自动记忆 文件：移除重复项、将规则迁移到适当的配置文件、压缩剩余条目、验证结果。典型效果：内存文件的令牌数量减少 30–50%。

自动记忆功能在 ~/.claude/projects/**/memory/MEMORY.md 中存储跨会话的上下文。启用：CLAUDE_CODE_DISABLE_AUTO_MEMORY=0 · 禁用：CLAUDE_CODE_DISABLE_AUTO_MEMORY=1

优势： 更快的上下文加载速度 · 无重复规则 · 更清晰的指令 · 更低的 API 成本

用法：

/memory-optimize          # 无参数 — 启动 4 步交互式工作流

技能文本专为 LLM 理解编写，并针对令牌效率进行了优化。

内存优化器

通过 4 个交互步骤优化 Claude Code 的内存文件。

广告位招租

在这里展示您的产品或服务

触达数万 AI 开发者，精准高效

联系我们

相关 Skills

find-skills 技能搜索工具 - Vercel Labs 开源智能体技能包管理器

705,000 周安装

Vercel React 最佳实践指南 | 58条Next.js性能优化规则与代码重构

245,300 周安装

Vercel Web界面规范检查工具 - 自动检测代码是否符合Web设计指南

196,800 周安装

agent-browser 浏览器自动化工具 - Vercel Labs 命令行网页操作与测试

128,699 周安装

React 组合模式指南：Vercel 组件架构最佳实践，提升代码可维护性

全局匹配所有内存文件：~/.claude/projects/**/memory/*.md
读取 ~/.claude/CLAUDE.md 和项目 CLAUDE.md（如果存在）
全局匹配 .claude/rules/*.md — 读取所有项目规则
读取 ~/.claude/rules/*.md — 读取所有全局规则

生成 Explore 代理以交叉引用所有已加载的文件
识别以下情况的条目：
- 相同规则已存在于 CLAUDE.md 中
- 相同模式已存在于规则文件中
- 与 CLAUDE.md 冲突（以 CLAUDE.md 为准）
显示分析结果：

发现 X 个重复/冗余条目（占内存的 Y%）： | 条目 | 内存文件 | 已存在于 | 操作 | |-------|-------------|------------|--------| | "首先使用 grepai" | MEMORY.md:5 | rules/grepai-first.md | 删除 | ...
AskUserQuestion："删除 X 个重复条目（占内存的 Y%）？这是安全的 — 内容已存在于其他地方。"
- 选项："是，全部删除" / "逐一审查" / "跳过此步骤"
如果获得批准，使用 Edit 工具应用删除操作

显示分类结果：

X 个条目适合迁移： | 条目 | 当前位置 | 目标 | 减少量 | |-------|-----------------|--------|-----------| | "始终使用 BD_PLUGIN_ROOT" | MEMORY.md:12 | .claude/rules/brewdoc.md | 15 个令牌 | ... 总计：X 个条目 → 约节省 Y 个令牌
AskUserQuestion："将 X 个条目迁移到 rules/CLAUDE.md？"
- 选项："是，全部迁移" / "逐一审查" / "跳过此步骤"
如果获得批准：
- 通过 Edit 创建/追加到目标规则文件
- 通过 Edit 从内存中移除已迁移的条目
- 如果目标文件不存在，则创建它

## 内存优化完成

### 摘要
| 指标 | 优化前 | 优化后 | 节省 |
|--------|--------|-------|-------|
| 总条目数 | X | Y | Z |
| 重复条目数 | X | 0 | — |
| 迁移条目数 | — | — | X |
| 令牌估计值 | ~X | ~Y | ~Z (~P%) |

### 所做更改
- 步骤 1：删除了 X 个重复条目
- 步骤 2：将 X 个条目迁移到 rules/CLAUDE.md
- 步骤 3：压缩了 X 个条目（减少 Y%）
- 步骤 4：修复了 X 个损坏的引用，移除了 X 个孤立的文件

### 最终内存结构
{~/.claude/projects/.../memory/ 的目录列表}

---

**brewdoc 的一部分：** [brewcode](https://github.com/kochetkov-ma/claude-brewcode) — 文档工具：内存优化、自动同步、Claude 安装文档、Markdown 转 PDF。
安装：`claude plugin marketplace add https://github.com/kochetkov-ma/claude-brewcode && claude plugin install brewdoc@claude-brewcode`

🇺🇸English

Plugin: kochetkov-ma/claude-brewcode

Memory Optimizer

Optimizes Claude Code auto-memory files in 4 interactive steps : removes duplicates, migrates rules to proper config files, compresses remaining entries, validates the result. Typical reduction: 30–50% token count in memory files.

Auto-memory stores context across sessions in ~/.claude/projects/**/memory/MEMORY.md. Enable: CLAUDE_CODE_DISABLE_AUTO_MEMORY=0 · Disable: CLAUDE_CODE_DISABLE_AUTO_MEMORY=1

Benefits: faster context loading · no duplicate rules · cleaner instructions · lower API cost

Usage:

/memory-optimize          # no args — starts 4-step interactive workflow

Skill text is written for LLM consumption and optimized for token efficiency.

Memory Optimizer

Optimizes Claude Code memory files through 4 interactive steps.

Nocontext: fork — must run in main conversation to spawn agents.

Phase 0: Load Context

Glob all memory files: ~/.claude/projects/**/memory/*.md
Read ~/.claude/CLAUDE.md and project CLAUDE.md (if exists)
Glob .claude/rules/*.md — read all project rules
Read ~/.claude/rules/*.md — read all global rules

Build context map:

memory_files: [paths]
claude_md_sections: [sections]
rules_files: [paths with content]

Step 1: Analysis — Remove Duplicates (Interactive)

Goal: Find memory entries that duplicate content already in CLAUDE.md or rules.

Spawn Explore agent to cross-reference all loaded files
Identify entries where:
- Same rule already in CLAUDE.md
- Same pattern already in a rules file
- Contradicts CLAUDE.md (CLAUDE.md wins)
Show analysis:

Found X duplicate/redundant entries (Y% of memory): | Entry | Memory File | Already In | Action | |-------|-------------|------------|--------| | "Use grepai first" | MEMORY.md:5 | rules/grepai-first.md | DELETE | ...
AskUserQuestion: "Delete X duplicate entries (Y% of memory)? This is safe — content exists elsewhere."
- Options: "Yes, delete all" / "Review each" / "Skip this step"
Apply deletion using Edit tool if approved

Step 2: Migration — Move to Rules/CLAUDE.md (Interactive)

Goal: Identify remaining memory entries better suited to persistent config files.

Decision tree (per entry):

Applies to ALL projects + IS a rule/constraint → ~/.claude/rules/
Applies to THIS project only + IS a rule → .claude/rules/
IS an architectural decision → project CLAUDE.md
IS a fact/pattern reusable across sessions → KEEP in memory

Show categorization:

X entries suitable for migration: | Entry | Current Location | Target | Reduction | |-------|-----------------|--------|-----------| | "Always use BD_PLUGIN_ROOT" | MEMORY.md:12 | .claude/rules/brewdoc.md | 15 tokens | ... Total: X entries → ~Y tokens saved
AskUserQuestion: "Migrate X entries to rules/CLAUDE.md?"
- Options: "Yes, migrate all" / "Review each" / "Skip this step"
If approved:
- Create/append to target rule files via Edit
- Remove migrated entries from memory via Edit
- If target file doesn't exist, create it

Step 3: Compression (Interactive)

Goal: Compress remaining entries using LLM-efficient formatting.

Compression techniques:

Prose → table row
Multiple related entries → single table
Verbose description → imperative one-liner
List of examples → pattern + one example

Show compression preview:

Compression opportunities found: | Before | After | Savings | |--------|-------|---------| | "When you need to... always use..." | "Use X for Y" | 8 tokens | ... Total: ~Y% token reduction (~Z tokens)

Show 2-3 specific before/after samples. 2. AskUserQuestion: "Compress remaining memory? (~Y% reduction)" * Options: "Yes, compress all" / "Skip compression" 3. Apply compression via Edit (bottom-up order to preserve line numbers)

Step 4: Validation (Automatic)

Goal: Verify final state and clean orphaned references.

Spawn reviewer agent to verify:
- No broken file path references in memory files
- No contradictions between memory and CLAUDE.md
- Memory files are well-formed markdown
Clean broken references (Edit tool)
Check for orphaned memory files (files in ~/.claude/projects/**/memory/ with no MEMORY.md reference)
Report orphaned files and ask to delete

Final Report:

## Memory Optimization Complete

### Summary
| Metric | Before | After | Saved |
|--------|--------|-------|-------|
| Total entries | X | Y | Z |
| Duplicate entries | X | 0 | — |
| Migrated entries | — | — | X |
| Token estimate | ~X | ~Y | ~Z (~P%) |

### Changes Made
- Step 1: Deleted X duplicate entries
- Step 2: Migrated X entries to rules/CLAUDE.md
- Step 3: Compressed X entries (Y% reduction)
- Step 4: Fixed X broken references, removed X orphaned files

### Final Memory Structure
{directory listing of ~/.claude/projects/.../memory/}

---

**Part of brewdoc:** [brewcode](https://github.com/kochetkov-ma/claude-brewcode) — docs tools: memory optimization, auto-sync, Claude installation docs, Markdown to PDF.
Install: `claude plugin marketplace add https://github.com/kochetkov-ma/claude-brewcode && claude plugin install brewdoc@claude-brewcode`

Weekly Installs

1.1K

Repository

kochetkov-ma/cl…brewcode

GitHub Stars

First Seen

Mar 1, 2026

Security Audits

Gen Agent Trust HubPass SocketPass SnykFail

Installed on

opencode1.0K

codex999

cursor998

amp973

github-copilot972

kimi-cli970

Claude Code 内存优化器：4步减少30-50%令牌，提升加载速度与API成本效益

🇨🇳中文介绍

内存优化器

内存优化器

相关 Skills

阶段 0：加载上下文

步骤 1：分析 — 移除重复项（交互式）

步骤 2：迁移 — 移至 Rules/CLAUDE.md（交互式）

步骤 3：压缩（交互式）

步骤 4：验证（自动）

🇺🇸English

Memory Optimizer

Memory Optimizer

Phase 0: Load Context

Step 1: Analysis — Remove Duplicates (Interactive)

Step 2: Migration — Move to Rules/CLAUDE.md (Interactive)

Step 3: Compression (Interactive)

Step 4: Validation (Automatic)