- Status:
active - Positioning: DevOps-style operations toolkit for AI coding agents
- Core Value: One command to sync, audit, and visualize multi-agent skill ecosystems
- Target Agents: Claude Code, Codex, Gemini CLI, Cursor, Antigravity
- Stack: Bash, Node.js,
vercel-labs/skills, local dashboard automation - Proof of Engineering: deterministic scripts, migration workflow, local-only control API
- Last Reviewed:
2026-03-02
AI Skills Control Kit is a production-ready local operations toolkit for managing agent skills across:
- Claude Code
- Codex
- Gemini CLI
- Cursor
- Antigravity
It provides:
- Deterministic CLI operations (
sync,status,dashboard,all) - Multi-agent skill synchronization through
vercel-labs/skills - Local dashboard generation for visibility and auditability
- Optional local web control API for one-click actions from the dashboard
- Portable install workflow for new machines
Most teams lose time and quality due to:
- Drift between agent skill folders
- Ad-hoc manual copying
- No visibility of coverage/missing skills
- Weak migration/recovery process
This kit standardizes those operations into one repeatable workflow.
.
├── install.sh
├── README.md
├── README.zh-CN.md
└── skills/
└── ai-skills-control/
├── SKILL.md
├── agents/openai.yaml
└── scripts/
├── sync_skills_all_agents.sh
├── build_skills_dashboard.mjs
├── skills_web_control_server.mjs
└── skills_manage.sh
- macOS/Linux shell environment
node >= 18npxavailablenpx -y skillscan run (fromhttps://github.com/vercel-labs/skills)- Local agent config folders exist (at least one of the supported agents)
chmod +x install.sh
./install.shThis installs runtime scripts into ~/scripts:
~/scripts/sync_skills_all_agents.sh~/scripts/build_skills_dashboard.mjs~/scripts/skills_web_control_server.mjs~/scripts/skills_manage.sh
# Full run: sync + dashboard + status
~/scripts/skills_manage.sh all
# Sync only
~/scripts/skills_manage.sh sync
# Rebuild dashboard only
~/scripts/skills_manage.sh dashboard
# View current global skills
~/scripts/skills_manage.sh status- Start local API server:
~/scripts/skills_manage.sh web- Open dashboard HTML and click buttons:
Run allRun syncRun dashboardRun status
Default dashboard path:
~/Documents/Obsidian Vault/20-Resources/工具/AI Skills 管理/skills-dashboard.html
- Canonical source of truth:
~/.agents/skills - Agent-specific folders are synced/linked for compatibility
- All execution is local-first and explicit via CLI commands
- Dashboard acts as an observability layer, not hidden automation
git clone https://github.com/outhsics/ai-skills-control-kit.git
cd ai-skills-control-kit
./install.sh
~/scripts/skills_manage.sh all
~/scripts/skills_manage.sh webnpx skillscommand fails:- verify Node/npm environment
- run
npx -y skills --version
- Dashboard not updated:
- run
~/scripts/skills_manage.sh dashboard
- run
- Buttons show server offline:
- run
~/scripts/skills_manage.sh web
- run
- A specific agent cannot see skills:
- run
~/scripts/skills_manage.sh status - then re-run
~/scripts/skills_manage.sh all - restart that agent
- run
- Web control API binds to
127.0.0.1only by default. - All commands executed by the API are allowlisted (
sync,dashboard,status,all). - This kit manages local files with user-level permissions; review scripts before use in regulated environments.
This project is intended for open-source collaboration and commercial operations. It is designed to be transparent, scriptable, and auditable for team workflows.
- Fork and create a branch
- Keep scripts deterministic and idempotent
- Update both language sections when changing behavior
- Open PR with before/after usage examples
MIT License. See LICENSE.
AI Skills Control Kit 是一套可用于生产环境的本地技能运维工具,用于统一管理以下 AI 工具的 skills:
- Claude Code
- Codex
- Gemini CLI
- Cursor
- Antigravity
它提供:
- 可重复执行的命令行运维入口(
sync、status、dashboard、all) - 基于
vercel-labs/skills的多工具同步 - 本地可视化看板(覆盖率/缺失项可见)
- 可选本地 Web 控制 API(在看板上按钮触发命令)
- 新电脑迁移安装流程
常见问题是:
- 多工具 skills 目录长期漂移
- 手工复制不可追踪、易出错
- 看不到哪些工具缺失某个 skill
- 换电脑恢复成本高
本项目把这些动作标准化为一条可审计、可复用的流程。
.
├── install.sh
├── README.md
├── README.zh-CN.md
└── skills/
└── ai-skills-control/
├── SKILL.md
├── agents/openai.yaml
└── scripts/
├── sync_skills_all_agents.sh
├── build_skills_dashboard.mjs
├── skills_web_control_server.mjs
└── skills_manage.sh
- macOS/Linux shell 环境
node >= 18- 可用
npx - 能执行
npx -y skills(来自https://github.com/vercel-labs/skills) - 本机已安装至少一个目标 AI 工具
chmod +x install.sh
./install.sh安装后会写入 ~/scripts:
~/scripts/sync_skills_all_agents.sh~/scripts/build_skills_dashboard.mjs~/scripts/skills_web_control_server.mjs~/scripts/skills_manage.sh
# 全流程:同步 + 刷新看板 + 状态检查
~/scripts/skills_manage.sh all
# 仅同步
~/scripts/skills_manage.sh sync
# 仅刷新看板
~/scripts/skills_manage.sh dashboard
# 查看当前全局技能状态
~/scripts/skills_manage.sh status- 启动本地控制服务:
~/scripts/skills_manage.sh web- 打开
skills-dashboard.html,点击按钮执行:
Run allRun syncRun dashboardRun status
默认看板路径:
~/Documents/Obsidian Vault/20-Resources/工具/AI Skills 管理/skills-dashboard.html
- 真源目录:
~/.agents/skills - 其它工具目录通过同步/软链接做兼容
- 所有动作命令化、显式执行
- 看板用于可视化观察,不做黑盒自动化
git clone https://github.com/outhsics/ai-skills-control-kit.git
cd ai-skills-control-kit
./install.sh
~/scripts/skills_manage.sh all
~/scripts/skills_manage.sh webnpx skills失败:- 检查 Node/npm 环境
- 执行
npx -y skills --version
- 看板数据不是最新:
- 执行
~/scripts/skills_manage.sh dashboard
- 执行
- 看板按钮显示 server offline:
- 执行
~/scripts/skills_manage.sh web
- 执行
- 某个工具看不到新 skill:
- 先
~/scripts/skills_manage.sh status - 再
~/scripts/skills_manage.sh all - 重启对应工具
- 先
- Web 控制服务默认只监听
127.0.0.1。 - 仅允许固定命令:
sync、dashboard、status、all。 - 本项目会操作用户本地目录;在合规场景建议先审阅脚本后再落地。
本项目面向开源协作与商业化运维场景,强调透明、可脚本化、可审计。
- Fork 后新建分支
- 保持脚本幂等、可重复执行
- 行为变更需同步更新中英文说明
- PR 中提供前后操作示例
MIT License,见 LICENSE。