How to Use Skills
This doc explains how to use CloudBase AI Skills in your project. Skills are “skill packs” that contain instructions and resources to inject CloudBase domain knowledge and step-by-step best practices into your AI.
When used together with MCP, the combination works best:
- MCP provides connection + permissions (so the AI can actually operate CloudBase)
- Skills provide rules + intuition (so the AI follows best practices)
Option 1: Install Skills (recommended)
Install CloudBase Skills into your local project via the AgentSkills spec. This works in AI tools that support Skills (e.g. Claude Code, Cursor, VS Code, GitHub Copilot, OpenCode, etc.).
Install
Run this in your project root:
npx skills add tencentcloudbase/cloudbase-skills
Follow the prompts to choose your AI tool. After installation, Skills will appear inside your project (e.g. .cursor/skills, .claude/skills, depending on the tool).
You can also tell your AI: “Install CloudBase Skills”.
Trigger Skills reliably
Skills may not always activate automatically (it depends on the model and the tool), but you can significantly improve activation through a few engineering approaches.
Skills use progressive loading: the AI first checks the skill name/description for relevance, and only loads the full instructions when needed.
To improve CloudBase skill activation:
- First-line injection: add a sentence at the start of your prompt, e.g. “You MUST read the cloudbase-guidelines skill FIRST when working with CloudBase projects.”
- Project-level rule: put the same constraint in
CLAUDE.mdorAGENT.mdin the project root so you don’t need to repeat it. - Hooks (editor hooks): use a pre-submit hook to force the AI to evaluate whether to use Skills before responding. This typically has the highest activation rate.
Use Skills with MCP
We recommend configuring CloudBase MCP as well:
- Connection modes (Local / Hosted) and setup steps: see Getting Started
- After MCP is configured in your IDE, install Skills as above so your AI can use both tools and skills in the same conversation.
Option 2: Manual copy
If you can’t use npx skills add (or you want full control), you can manually copy rules/prompts into your project and let your AI tool load them via “include file” features.
Where to copy and how to reference
- Rules: put Markdown rules somewhere your AI tool can read them, for example:
- Cursor:
.cursor/rules/or anywhere in the repo, then reference via@Files - Claude Code / others:
CLAUDE.md,AGENT.md, orprompts/*.md
- Cursor:
- Prompts: save a single prompt as
prompts/<name>.mdand reference it via your tool’s file include syntax.
Examples:
- GitHub Copilot:
#prompts/auth-web.md - Cursor: use
@Filesand select the file - Zed: use
/fileand select the file
Available CloudBase prompt scenarios are listed below.
Activation summary
| Option | Pros | Typical activation |
|---|---|---|
| Option 1: Install Skills | easy, works great with MCP; can be boosted via injection/rules/hooks | low by default; much higher with injection/project rules; highest with hooks |
| Option 2: Manual copy | flexible and fully customizable; depends on explicit file references | reliable when you explicitly reference the file |
Available Skill Scenarios
身份认证
数据库
Further Reading
- MCP: Connection modes, Getting Started
- Tutorials: Videos & tutorials