Skill: How to Reuse a Way of Doing Things

Skill: Organize the methods, rules, and supporting materials for a type of task into a reusable package for the Agents that support it to use as needed.

Here, an Agent refers to an AI system that can advance tasks around a goal and decide subsequent steps based on process results. Even without diving into its internal mechanisms for now, you can understand how a package of methods helps reuse approaches.

If every time you prepare materials you have to repeat "first check student basics, arrange activities for 40 minutes, mark where materials are insufficient, finally review the finished product," you can keep these common practices and attach a lesson plan template and review instructions.

The next time, you still need to specify the topic, grade level, and materials. A Skill helps reuse methods, but it doesn't decide your current goal for you.

What Belongs in a Skill

Using the example of repeatedly preparing classroom materials:

For a one-time, simple request, a prompt alone is sufficient. Tasks that are repeated often, easy to miss steps in, or have fixed materials are better organized into a reusable method package.

How It Differs from a Prompt

Prompt expresses this specific request: "Prepare a Silk Road lesson plan for first-year middle school students."

Skill describes how to handle this type of task: "First confirm student basics and goals, then organize activities, finally review materials."

The two work together. Copying a commonly used passage repeatedly is also reusing methods; adopting the Skill format also lets you organize instructions, templates, scripts, and discovery conditions. Whether it gets automatically discovered and how it's activated depends on the application that supports it.

Before Using, Check Three Things

First see if it's suitable for the current task, then check what capabilities are needed, and finally verify the actual results.

For example, if a method package says "generate PPT then view screenshots of each page," the environment must have tools for generating files and rendering pages; if the AI also needs to inspect the pages, it must have corresponding image reading and judgment capabilities. When these capabilities are missing, the method package should specify where to stop and what can be completed.

After use, you can verify: whether it actually read the required materials; whether it organized according to the template; whether it executed the applicable checks. Don't judge effectiveness based solely on "used the so-and-so Skill."

Going a Bit Deeper: What's in the Folder

In the open Agent Skills format, a Skill is a folder containing SKILL.md. This file records the name, usage instructions, and specific methods, and may also include scripts, reference materials, and templates.

The following is only a structural illustration and has not been installed in any application:

lesson-preparation/
  SKILL.md              When to use, how to prepare lessons, when to stop
  references/           Material usage and review instructions
  assets/               Lesson plan templates, presentation styles
  scripts/              Optional timing statistics or file inspection programs

Applications typically first read the name and brief description, load the full instructions upon finding a match, and read other materials as needed. This approach reduces the amount of information loaded into context at the start. How discovery, loading, and execution work in specific clients should be verified against their documentation.

Scripts are programs; whether they can run still depends on tools, environment, and permissions. A method package saying "files can be modified" doesn't mean the system already permits it to modify any file.

Three Commonly Confused Points

• Loading a Skill is not retraining the model. The common mechanism is providing the methods and related materials to the Agent for reference.
• Writing operational steps doesn't mean the corresponding capabilities exist. File generation, network access, and image reading still require tool support.
• Packaging methods doesn't make results inherently reliable. The methods may not suit the current materials, or execution may fail—review is still needed.

MCP addresses how applications communicate with external services; Skill addresses how reusable methods and materials are organized. They can work together but serve different purposes. See Comparison of Four Terms.

Think About It

A "organize course announcements" Skill specifies extracting course name, time, location, and deadline. The announcement doesn't state a deadline—should the method package require adding "tomorrow afternoon"?

Reference judgment: Mark as not provided, confirm when necessary. Reliable methods should also specify how to handle missing items, not just define a polished output format.

Next Steps

To express this specific task clearly, read Prompt; to understand how external materials connect, read Model Context Protocol (MCP).

Sources and Scope

Referenced Agent Skills Overview and Specification, verified: 2026-09-09. The file structure and on-demand loading explanations in this article are limited to that open format; identically named features in different products may differ. The structural illustration was not tested for cross-application installation and does not serve as a working product installation guide.