Skip to main content

Basic Concepts

To get started with AI-assisted development, first get to know three concepts: large language models, AI agents, and skills.

Large language model​

  • Role: Understand questions and generate responses.

A large language model powers many AI tools. It learns from large amounts of data and uses the information you provide to explain concepts, analyze problems, and write code. You can communicate with it in everyday language.

A large language model predicts and generates content based on context and probabilities. Even a response that sounds reasonable may contain errors and needs human review and verification.

However, generating code does not mean the development work is complete. The model itself cannot directly open your project, edit files, or run a build. These actions require tools.

Reference: What Are Large Language Models (LLMs)?.

AI agent​

  • Role: Combine a large language model with tools to perform tasks.

An AI agent combines a large language model with tools. You describe a goal, and it can break the task into steps, read the project, modify code, and make further adjustments based on execution results.

For example, if you want to add a button to a plugin, a large language model can explain how to write the code. An agent with the necessary tools and permissions can make changes directly in the project and attempt to verify them by running a build.

You define the goal and confirm the results; the agent helps carry out the work. An agent can also make mistakes when interpreting tasks, calling tools, or performing actions. Its final output still needs human review and validation.

Reference: What Are AI Agents?.

Skill​

  • Role: Capture reusable knowledge and guide an agent through tasks.

When you repeatedly ask an agent to do the same kind of work, you can organize the common steps and considerations into a skill.

A skill is a reusable set of task instructions for an agent. It can also include scripts, templates, and reference materials. With a skill, the agent can follow an established process without you having to explain it each time.

A skill folder has the following structure. Scripts, reference materials, and templates can be added as needed:

meeting-notes/
├── SKILL.md # Required: skill instructions
├── scripts/ # Optional: helper scripts
├── references/ # Optional: reference materials
└── assets/ # Optional: templates and other resources

A minimal skill needs only a folder containing a SKILL.md file. Here is a simplified SKILL.md example for preparing meeting minutes:

---
name: meeting-notes
description: Use when the user wants to turn meeting notes into meeting minutes.
---

1. Read the meeting notes provided by the user and extract the key discussion points and confirmed conclusions.
2. List action items, owners, and deadlines. Mark any information not provided as "To be confirmed".
3. Organize the output into three sections: Discussion Points, Meeting Conclusions, and Action Items.

Here, name identifies the skill, description explains when to use it, and the body tells the agent what to do.

Reusable skills are available for Rokae+ plugins. See Rokae+ Skills.

Reference: Agent Skills Overview.