reflect-and-remember
反思记忆协议
反思自问清单（完成任务后快速过一遍）
- 这次有没有踩坑/走弯路？根因是什么？
- 有没有反直觉的行为（框架、API、工具的意外表现）？
- 有没有架构决策或项目约定值得记录？
- 这些信息下次还会用到吗？其他开发者会受益吗？
全部为否 → 跳过，不记忆。
存储位置
| 类别 | 位置 | 进 git | 内容 |
|---|---|---|---|
| 团队共享 | 项目 .claude/memory/MEMORY.md |
是 | 部署教训、业务规则、架构决策、环境要点、踩坑记录 |
| 敏感私有 | ~/.claude/projects/<project>/memory/MEMORY.md |
否 | 密码、密钥、token、个人路径 |
写入流程
- Read 目标文件，检查是否已有相关记忆
More from nangongwentian-fe/agent-skills
persistent-memory
通用跨会话记忆协议（Universal Memory Protocol）。让所有 AI 编程工具共享同一套记忆系统。适用 Claude Code / Cursor / Aider / Cline / Codex / Trae / OpenCode。能力：智能分类 / FSRS 衰减 / 月度压缩 / 多层检索。触发：用户说"记住"；问"之前"；检测到敏感信息；会话结束。
23exa-unified-research
Use this skill for ANY web research task — looking up companies, finding people, searching for code examples or API docs, reading tech blogs, finding academic papers, checking SEC filings or financial reports, or gauging social media sentiment. Covers any question that requires searching the internet for current information. Exa provides neural/semantic search that returns higher-quality, better-structured results than simple keyword search. Load this skill whenever the user asks you to search, research, find, look up, or investigate anything online — even if you think you can handle it with built-in search tools, this skill will produce significantly better results.
10git-rebase-workflow
Git Rebase 分支同步流程，用于将当前功能分支 rebase 到最新的目标分支（如 master/main），保持提交历史整洁。适用于功能分支落后于目标分支时，需要同步最新代码的场景。
8code-review-uncommitted
对 git 中未提交的代码变更进行多维度 code review，包括项目规范合规性、Bug 扫描、代码注释合规性、组件封装/架构设计合理性审查，并通过置信度评分过滤误报。当用户要求 review 未提交的代码、review 当前改动、或使用 /code-review-uncommitted 时触发。
8codex-cli
Delegate tasks to the local OpenAI Codex CLI (`codex exec`, `codex review`, `codex cloud exec`). Use this skill whenever the user asks to run Codex, delegate a task to Codex, use OpenAI's agent, do a Codex review, or mentions 'codex' in any form — including phrases like '用 codex 跑一下', '让 codex 帮忙', 'run this with codex', 'codex exec', 'codex review', 'send to codex cloud'. Also trigger when the user wants a second opinion from another AI agent on code changes, or wants to run a task in a sandboxed OpenAI environment.
1web-content-fetcher
>
1