Skills Tutorial

Why Agent Skills are needed

Although AI agents have powerful general capabilities, they often lack the necessary context and professional knowledge when handling specific domain tasks.

For example, an AI agent may not know what coding standards your company uses, may not understand how to use a specific API, or may be unclear about the specific steps of a business process.

The core idea of Skills is:Encapsulate human professional knowledge into skill packs, allowing AI agents to automatically load and use them when needed.

What Agent Skills can bring

Capability	Description
Domain expertise	Encapsulate domain-specific knowledge (such as legal review processes, data analysis pipelines) into reusable instructions and resources
Repeatable workflows	Turn multi-step tasks into consistent, auditable standard processes
Cross-product reuse	Write a Skill once and use it in any tool that supports the Agent Skills format

Common problems when there are no Skills

Without Skills, agents may repeatedly ask the same clarifying questions, miss team-specific naming conventions, and forget important edge cases.

With Skills, these experiences can be written directly into files, and the agent will follow them every time.

The relationship between Agents and Skills

An AI agent is an AI system that can perceive, reason, and act. It doesn't just chat; it can also execute tasks.

Skills are instruction manuals (Markdown files) that tell agents how to complete certain types of tasks. The value of Skills is making task execution accurate, stable, and reusable.

智能体的大模型 = 大脑（负责理解、决策、说话）
Skills = 手和脚（负责动手干活）

Imagine that when you first join a company, you will have two completely different AI assistants:

Assistant A (ordinary conversational AI):

• One question, one answer; passive response.
• If you ask 'How to write a quarterly report', it will only give you a generic template.
• It can't open spreadsheets, can't query data on its own, and can't organize files, integrate, or send.
• In essence, it's just a talking online encyclopedia—it gives answers but doesn't execute.

Assistant B (AI agent)

• Direct commands, fully automated closed-loop execution.
• You only need one sentence: organize this quarter's sales data, generate a report, and send it to General Manager Zhang.
• It can autonomously complete the entire process: retrieving files, reading raw data, automatically generating charts, writing complete reports, and sending emails with one click.
• No manual operation is needed throughout the process; just sit back and wait for the results.

Three core capabilities of Agents

Capability	Description	Example
Perception	Receive input, understand context	Read the files you upload, understand your needs
Reasoning	Make plans, decide what to do	Determine 'first read the file, then generate the report, and finally send the email'
Action	Call tools, execute steps	Actually open files, write content, and send it out

---

Why are Skills needed? What problems do they solve?

Ordinary AI agents (such as Claude or DeepSeek) are smart, but they can easily make mistakes when lacking specific context. For example:

• The team has its own coding standards, but the AI needs to be manually reminded every time.
• For complex processes like handling PDF forms or debugging GitHub Actions, the AI may not know the best practices.

Agent Skills solve these problems:

• Automatic triggering: AI automatically loads relevant skills based on the task, with no need to manually input long prompts.
• Reusable & shareable: Created once, used by the whole team or community, with Git version control support.
• Efficient context utilization: Uses progressive disclosure, loading only the necessary parts to avoid context window overflow.
• Cross-platform: The same Skill can be used in tools such as Claude, VS Code Copilot, Cursor, etc.

Other extensions