Multi-platform Agent Skills monorepo that mounts as a DeepSeek Harness profile bundle, exposing Chinese writing, Git, plugin-dev, taste, and semantic-linter skills.
Install
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:SummerSec/SumSec-Skills
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time. Only install sources you trust, and pin a commit (github:owner/repo#sha).
README
SummerSec 个人 Agent Skills 集合，按类别分插件管理。
仓库布局
SumSec-Skills/
├── writing-zh/ # 中文写作插件
│ └── skills/
│ ├── humanizer-zh/ 去 AI 味润色
│ └── sumsec-illustrations/ SumSec 博客正文配图
├── dev-tools/ # 开发工具插件
│ └── skills/
│ ├── git-commit-pr/ Git 提交与 PR
│ ├── agent-chat-history/ 对话历史检索
│ └── frontend-design/ 前端界面实现
├── agents-dev/ # Agent 开发生态插件（聚合）
│ └── skills/
│ ├── skill-creator/ 技能创建（claude-plugins-official）
│ ├── writing-rules/ Hook rules 生成（hookify）
│ ├── agent-sdk-dev/ Agent SDK 开发
│ ├── claude-agents-symlink/ CLAUDE.md 指向 AGENTS.md
│ ├── skill-optimizer/ Skill 审计优化
│ ├── multi-platform-plugin-guide/ 多平台版本对齐
│ └── workflow-skill-creator/ 流程编排 Skill 设计
├── skills/ # 通用 skill 聚合入口（symlink 到各插件 skills）
├── plugin-dev/ # 镜像：插件开发七件套（agent/command/hook/skill/MCP/structure/settings）
├── claude-md-management/ # 镜像：CLAUDE.md 维护
├── hookify/ # 原创：Hook 创建工具
├── cloudflare-email/ # Cloudflare 临时邮箱插件
├── taste-skill/ # Taste Skill 跨平台插件封装
├── semantic-linter/ # Semantic-Linter 跨平台插件封装
├── openclaw.plugin.json # OpenClaw 插件清单
├── openclaw/ # OpenClaw 插件入口 & skills
├── opencode/ # OpenCode 插件入口 & rules
├── hermes/ # Hermes skills & context
├── dsh/ # DeepSeek Harness bundle patch & 文档
├── .claude-plugin/ # 根 marketplace
├── .cursor-plugin/
├── .codex-plugin/
├── .agents/plugins/
├── .cursor/rules/
├── taste-skill-upstream/ # submodule: Leonxlnx/taste-skill
├── semantic-linter-upstream/ # submodule: SummerSec/semantic-linter
├── AGENTS.md
├── README.md
├── package.json
└── plugin.json
安装
Claude Code
/plugin marketplace add https://github.com/SummerSec/SumSec-Skills.git
/plugin install writing-zh@sumsec-skills
/plugin install dev-tools@sumsec-skills
/plugin install agents-dev@sumsec-skills
/plugin install plugin-dev@sumsec-skills
/plugin install claude-md-management@sumsec-skills
/plugin install hookify@sumsec-skills
/plugin install cloudflare-email@sumsec-skills
/plugin install taste-skill@sumsec-skills
/plugin install semantic-linter@sumsec-skills
OpenAI Codex
codex plugin marketplace add SummerSec/SumSec-Skills --ref master
然后在 Codex 的 /plugins 里选择 SumSec Skills，按需安装 writing-zh、dev-tools、agents-dev、plugin-dev、claude-md-management、hookify、cloudflare-email、taste-skill、semantic-linter 等插件。仓库内的 .agents/plugins/marketplace.json 是 repo-scoped marketplace，每个插件条目使用 source: "local" + 对应子目录 path（如 ./writing-zh）指向当前 checkout；Git 获取由 codex plugin marketplace add 管理，不在插件条目里写 url/ref。
Codex 也会直接扫描仓库级 .agents/skills/，适合放仅服务本仓的工具型 skill；需要分发给其他项目或团队时，应打包进 .codex-plugin/plugin.json 所描述的插件。
DeepSeek Harness
临时从当前 checkout 加载全部插件 Skill：
dsh --profile headless --patch ./dsh/cordis.patch.yml "使用合适的 SumSec Skill 检查当前项目"
持久启用时，先把本仓库安装到目标 profile，再将包名 sumsec-skills 追加到该 profile 的 dsh.profile.bundles。完整步骤、Git 安装方式与 prerelease 注意事项见 dsh/README.md。
手动安装（软链接）
将 <plugin>/skills/<skill-name>/ 链接到对应客户端 skill 目录：
ln -sf "$(pwd)/dev-tools/skills/git-commit-pr" ~/.claude/skills/git-commit-pr
| 客户端 | skill 目录/安装方式 |
|---|---|
| Claude Code | /plugin install <plugin>@sumsec-skills |
| Cursor | .cursor-plugin/marketplace.json 导入 |
| OpenAI Codex CLI | codex plugin marketplace add SummerSec/SumSec-Skills --ref master |
| DeepSeek Harness | dsh --profile headless --patch ./dsh/cordis.patch.yml 或 profile bundle |
| OpenClaw | openclaw.plugin.json + openclaw/ 插件加载 |
| OpenCode | opencode/plugins/sumsec-skills.mjs 插件注册 |
| Hermes | hermes/skills/sumsec-skills/SKILL.md 复制加载 |
| 通用 symlink | ~/.agents/skills/<name>/ -> <plugin>/skills/<name>/ |
技能一览
writing-zh（中文写作）
| 技能 | 来源 | 说明 |
|---|---|---|
| humanizer-zh | 本仓库 | 去 AI 味：本地 CLI + 深度指南，反 AI 审查二遍工作流 |
| sumsec-illustrations | 本仓库 | 为 sumsec.me 风格文章生成 SumSec Observer 正文配图 |
dev-tools（开发工具）
| 技能 | 说明 |
|---|---|
| git-commit-pr | 安全完成 commit、push、PR/MR |
| agent-chat-history | 按日期查本机 Agent 历史对话 |
| context7-cli | context7 CLI：查询库文档 |
| context7-mcp | context7 MCP 服务器集成 |
| find-docs | 查找库文档（context7） |
| frontend-design | 打造高质量前端界面、页面和应用 |
agents-dev（Agent 开发生态）
| 技能 | 来源 | 说明 |
|---|---|---|
| skill-creator | claude-plugins-official | 技能创建全流程 |
| writing-rules | hookify | Hook 编写与 rules 生成 |
| agent-sdk-dev | claude-plugins-official | Agent SDK 开发 |
| claude-agents-symlink | 本仓库 | 统一项目根 CLAUDE.md/claude.md 软链接到 AGENTS.md/agents.md，适配 Windows/macOS/Linux |
| skill-optimizer | 本仓库 | Skill 审计优化（路径 A 改 / 路径 B 只读八维） |
| multi-platform-plugin-guide | 本仓库 | 多平台版本对齐与发布清单 |
| workflow-skill-creator | 本仓库 | 复杂流程编排 Skill 设计 |
插件开发七件套（agent/command/hook/skill/MCP/structure/settings）请安装
plugin-dev，已不在agents-dev内重复维护。
plugin-dev（插件开发七件套）
| 技能 | 说明 |
|---|---|
| agent-development | Agent 开发 |
| command-development | 命令开发 |
| hook-development | Hook 开发 |
| mcp-integration | MCP 集成 |
| plugin-settings | 插件设置 |
| plugin-structure | 插件结构 |
| skill-development | 技能开发 |
附带 3 个 agents（agent-creator、plugin-validator、skill-reviewer）和 /plugin-dev:create-plugin 引导式工作流。
其他镜像插件
| 插件 | 说明 |
|---|---|
| claude-md-management | 维护和改进 CLAUDE.md：质量审计 + 会话学习捕获 |
| hookify | 通过对话模式分析创建 hooks，支持正则匹配与多事件类型 |
| cloudflare-email | 通过 Address JWT 读取、获取和发送 Cloudflare 临时邮箱邮件 |
| taste-skill | 来自 Leonxlnx/taste-skill 的 13 个前端设计、重设计、image-to-code、品牌与视觉方向 Skill |
| semantic-linter | 检测 Skill/Prompt/Agent 指令中的宽边界用词；含 hooks、CLI 扫描、词典与项目规则安装 |
semantic-linter（语义边界检测）
| 技能 | 说明 |
|---|---|
| semantic-analyzer | 词典之外的深度语义审阅 |
| semantic-linter-shot | 单文件宽边界词速查 |
| lexicon-manager | 维护语义陷阱词典 |
| rules-installer | 安装项目级 semantic-rules 与托管指令块 |
Git Submodule 与 Skill 同步
本仓库通过 submodule 引用第三方 skill，用同步脚本复制到插件目录（替代 symlink）。
| Submodule | 来源 | 提供的 skill |
|---|---|---|
| claude-plugins-official | anthropics | skill-creator, frontend-design, plugin-dev 系列, hookify, agent-sdk-dev, claude-md-management |
| context7 | upstash | context7-cli, context7-mcp, find-docs |
| taste-skill-upstream | Leonxlnx | taste-skill 插件的 13 个设计 Skill |
| semantic-linter-upstream | SummerSec | semantic-linter 插件（skills/hooks/commands/lib/bin） |
# 新机器安装
git clone --recurse-submodules https://github.com/SummerSec/SumSec-Skills.git
cd SumSec-Skills
python .claude/skills/sync-skills/scripts/sync-skills.py
许可
Apache-2.0
Ponytail in dev-tools
The dev-tools plugin bundles Ponytail-focused skills:
dev-tools/skills/ponytail/dev-tools/skills/ponytail-review/dev-tools/skills/ponytail-audit/dev-tools/skills/ponytail-help/
Links
More in this category
GanyuanRan/Aegis★ 1023
Software-engineering method pack for coding agents, with skills for baseline-first planning, systematic debugging, prompt hygiene, verification before completion, and repair/retirement tracking.
superdesigndev/superdesign-skill★ 417
Design skill for UI and marketing graphics on the Superdesign canvas: reads the repo for context, extracts its design system, then generates and iterates branchable design drafts, flow pages, and reusable components through the Superdesign CLI.
Ikalus1988/MisakaNet★ 389
Failure-recovery memory: search and record failure-recovery lessons from real engineering sessions, with BM25 + semantic RAG retrieval and a lessons knowledge base.
Vladimir-Human/humanizer-ru#dsh★ 100
Cleans AI traces from Russian text: finds chatbot copy-paste artefacts (ChatGPT, Gemini, Grok, Perplexity, DeepSeek) and rewrites the text into natural prose on request; 39 regex markers with an evidence registry, offline, text-only bundle.
dhicoc/dsh-reverse-skill★ 18
Complete reverse-skill pack (85 SKILL.md) as a DeepSeek Harness Cordis plugin: reverse engineering, authorized pentesting and security-research skill router.
zhaiyateng/dsh-design-skills★ 10
Design-aesthetics skill pack (10 styles: dark SaaS, minimal white, neumorphism, brutalism, glassmorphism, Japanese minimal, bento grid, cyberpunk, vaporwave, art deco) with runnable landing-page demos: tokens, component rules, forbidden lists, and acceptance checklists per style.