claude-skills-zh
CLAUDE.md2 starsChanged 5 months agoMITView on GitHub
@~/.openclaw/workspace/MEMORY.md 当你需要了解用户偏好、项目上下文、历史决策时,先读取上述文件。 如果该文件存在但内容为空,正常继续。 自动发布: 每次对记忆系统/memory/skills 的改进,自动 commit + push 到上述仓库。
Files
CLAUDE.md — Persistent Memory for Claude Code
跨工具共享记忆
@~/.openclaw/workspace/MEMORY.md
当你需要了解用户偏好、项目上下文、历史决策时,先读取上述文件。 如果该文件存在但内容为空,正常继续。
你是谁
- Windows 用户,中文为主 | Python/PowerShell | 目标:AI 长期记忆,不"失忆"
记忆系统
本文件 + self-improving/ 目录 = 你的长期记忆系统。
文件结构
self-improving/memory.md— 热记忆(偏好、模式、规则)→ 每次会话必读self-improving/corrections.md— 纠错索引(详情见~/.claude/LEARNED.md)self-improving/heartbeat-state.md— 心跳状态(自动维护)self-improving/heartbeat-rules.md— 心跳规则(检查间隔/阈值/告警)self-improving/session-handoff/— 跨会话交接文档(Pro Workflow 生成)self-improving/projects/— 各项目专属记忆self-improving/domains/— 领域知识self-improving/archive/— 归档的旧记忆~/.claude/LEARNED.md— Pro Workflow 学习日志(learn-rule / replay-learnings)~/.openclaw/workspace/MEMORY.md— 跨工具共享记忆(含"最近对话"时间线)C:\Users\xnsrz\CLAUDE.md— 入口(本文件)
会话流程
- 会话开始时:读取本文件 → 读取
~/.openclaw/workspace/MEMORY.md(含最近对话时间线)→ 读取self-improving/memory.md→ 读取self-improving/corrections.md - 会话中:如果用户纠正我的错误 → 自动追加到
~/.claude/LEARNED.md(corrections.md只更新索引条目) - 会话中:如果用户表达了偏好/规则 → 自动更新
memory.md - 会话结束时:用户可以用
/memory命令保存重要上下文;重要会话执行 wrap-up → 生成 session-handoff → 更新 MEMORY.md"最近对话"
工具
- MIMO Vision(图片/视频理解)| Python 脚本:gen_img / gen_vid / mimo / multimedia
- CloakBrowser(反检测浏览器,过 Cloudflare/reCAPTCHA)|
cloakbrowser_tool.py - stealth_browser(4层反反爬:L1 nodriver → L2 指纹轮换 → L3 rebrowser → L4 CloakBrowser)
插件
- Pro Workflow(记忆/学习/hooks)| Superpowers v5.1(TDD/调试/子代理/14 skills)| CodeGraph(代码知识图谱)
- Academic Research Skills v3.9(学术研究4件套:论文/审稿/管道/深度研究)
- Matt Pocock Skills(11万⭐, 24个真实工程师技能:TDD/诊断/原型/PRD/issue管理等)
- Claude Code 中文插件 v2.4.35(112文件,状态栏/提示简体中文)
环境
- Windows | PowerShell |
bypassPermissions
Context 工程(Pro Workflow: context-optimizer + compact-guard + context-engineering)
- Write — 重要状态写 NOTES.md / memory.md,而非依赖上下文窗口
- Select — Grep 定位 → offset+limit 定点读,永远不 dump 整个大文件
- Compress — 50% 即触发
/compact;compact 前保存 5 文件清单 + 当前任务+"下一步" - Isolate — 重搜索/测试→subagent,主 session 保持干净
- 预算: 规划<20% / 实现<50% / 测试<70% / 审查<85%
- Auto-compact: 建议
CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=50
规则
- 优先使用中文回复
- 修改文件前先确认,但简单操作可直行
- 如果学到用户的偏好/习惯,主动记录到
self-improving/memory.md - 如果犯错被纠正,主动记录到
~/.claude/LEARNED.md(corrections.md 只做索引) - 每 5 次会话检查一次
heartbeat-state.md,看记忆系统是否健康 - MEMORY.md"最近对话"保留最近 5 条,超量自动归档到
self-improving/archive/conversation-timeline.md - 重要会话结束:执行 wrap-up → 生成 session-handoff → 更新 MEMORY.md 速查+最近对话
- 高效输出:不谄媚(无"Sure!/Great!"),不啰嗦(无"Let me know if..."),不复述问题,代码优先解释在后
- 结构化优先:表格/列表 > 段落散文,ASCII字符 > Unicode装饰符
- 记忆系统优化后 → 同步推送到 Git:
https://github.com/0mlkrizzz655335v/Claudecodeskills-
自动发布: 每次对记忆系统/memory/skills 的改进,自动 commit + push 到上述仓库。
已学教训(详情见 ~/.claude/LEARNED.md)
[LEARN] Context: 装完插件≠生效,新工具安装后验证产出文件 [LEARN] Configuration: 用户偏好优先用全局持久化而非每会话命令 [LEARN] Memory: 纠错统一入口 LEARNED.md,corrections.md 仅做索引 [LEARN] Memory: 文件格式决定工具能否检索 — 必须用 [LEARN] 标签
agentmemory 桥接区
<!-- 此区域由 agentmemory CLAUDE_MEMORY_BRIDGE 自动维护,勿手动编辑 --><!-- LINE_BUDGET=80,超出后 agentmemory 自动裁剪最旧内容 --><!-- AGENTMEMORY_BRIDGE_START --><!-- AGENTMEMORY_BRIDGE_END -->In this repository57 other files it gives its agents.
- codex-skills-pack/AGENTS.mdAGENTS.md
- automateSkill
- autopilotSkill
- create-hookSkill
- create-subagentSkill
- goalSkill
- loopSkill
- onboardSkill
- subscribeSkill
- review-agentSkill
- review-bugbotSkill
- review-securitySkill
- reviewSkill
- split-to-prsSkill
- canvasSkill
- migrate-to-buildsSkill
- migrate-to-skillsSkill
- new-repoSkill
- originSkill
- rename-chatSkill
- sdkSkill
- shareSkill
- shellSkill
- statuslineSkill
- update-cli-configSkill
- update-cursor-settingsSkill
- computer-useSkill
- documentsSkill
- excel-live-controlSkill
- google-docsSkill
- google-drive-commentsSkill
- google-driveSkill
- google-sheetsSkill
- google-slidesSkill
- pdfSkill
- PresentationsSkill
- sites-buildingSkill
- sites-hostingSkill
- sites-preview-troubleshootingSkill
- SpreadsheetsSkill
- template-creatorSkill
- create-ruleSkill
- create-skillSkill
- imagegenSkill
- openai-docsSkill
- plugin-creatorSkill
- plugin-managementSkill
- skill-creatorSkill
- skill-installerSkill
- humanizer-zh-academicSkill
- natural-fictionSkill
- natural-writingSkill
- precise-proseSkill
- prose-auditSkill
- qu-ai-weiSkill
- research-paper-writingSkill
- voice-profileSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

