OpenClaw Glossary: Every Term You Need to Know

OpenClaw has its own vocabulary, and if you are new to the ecosystem the terminology can feel overwhelming. This glossary defines every important term in plain language so you can follow along with documentation, tutorials, and community discussions without getting lost.

Bookmark this page. You will come back to it.

Agent

An agent is an AI system that can take actions on your behalf. In the OpenClaw context, the agent is the AI coding assistant that runs in your terminal. It reads your code, responds to your prompts, writes files, runs commands, and uses tools. Unlike a simple chatbot that only generates text, an agent can interact with your environment — your filesystem, your shell, your development tools.

When people say "my agent did X," they mean the OpenClaw AI performed that action during a session.

Bazaar

The Bazaar is the community marketplace for OpenClaw skills. You can browse it at the skills directory. It hosts over 2,300 skills that developers have created and shared. Every skill is open source, rated by the community, and free to install. Think of the Bazaar as an app store specifically for OpenClaw agent capabilities.

Context Window

The context window is the total amount of text your agent can process at once. It includes the conversation history, any files it has read, installed skill instructions, and system prompts. When the context window fills up, older information gets dropped.

Why this matters: if you install too many skills or load too many files, you use up context window space that could go toward understanding your actual question. Managing your context window effectively is part of getting good results.

Hooks

Hooks are automated actions that trigger at specific points during an OpenClaw session. For example, you can set up a hook that runs a linter after every code change, or a hook that logs every agent action to a file. Hooks let you build workflows around your agent without manual intervention.

Hooks are configured in your settings.json file. They follow an event-based model: you specify when the hook should fire and what command it should run.

MCP Server

MCP stands for Model Context Protocol. An MCP server is a lightweight service that gives your agent access to external tools and data sources. For example, an MCP server might connect your agent to a database, a project management tool, a design system API, or a documentation source.

When you configure an MCP server, your agent gains new abilities beyond reading files and running shell commands. It can query your database, look up Jira tickets, fetch Figma designs, or interact with any service the MCP server exposes.

How MCP Servers Differ From Skills

Skills provide instructions — they tell your agent how to behave. MCP servers provide tools — they give your agent new actions it can take. A skill might say "always use parameterized queries." An MCP server might give the agent the ability to actually run a query against your database.

Memory

Memory is the system that lets your agent remember information across sessions. Without memory, every OpenClaw session starts fresh — the agent has no recollection of what you discussed yesterday or what decisions you made last week.

With memory enabled, the agent can store and retrieve facts, preferences, and decisions. For example, it might remember that your project uses tabs instead of spaces, that your API follows a specific naming convention, or that you prefer functional components over class components.

Memory is stored locally in your project. You control what gets remembered and can clear it at any time.

Model

The model is the large language model (LLM) that powers your agent. OpenClaw supports multiple models from different providers. Each model has different strengths — some are faster, some handle longer contexts better, some are better at reasoning through complex problems.

You can switch models in your configuration. The default model works well for most tasks, but experienced users sometimes switch models depending on the task at hand.

Persona

A persona is a pre-configured bundle that includes a system prompt, a set of curated skills, and sometimes specific model and tool settings. Personas are designed for specific roles or workflows. A "Frontend Engineer" persona might come with React, TypeScript, and testing skills pre-installed, while a "DevOps Engineer" persona might include Terraform, Docker, and CI/CD skills.

Personas save you the time of manually selecting and installing individual skills from the Bazaar. They are a great starting point, especially for beginners who are not sure which skills to pick.

Plugin

In the OpenClaw ecosystem, "plugin" is an older term that has largely been replaced by "skill." You may still see it in older documentation and community posts. When someone mentions an OpenClaw plugin, they almost always mean a skill. The functionality is the same — a modular set of instructions that extends your agent's capabilities.

Project Configuration

Project configuration refers to the settings stored in your project's .openclaw/ directory. This includes installed skills, agent preferences, memory data, and any custom settings you have defined. Project configuration is local to each project, so your Python project can have different skills and settings than your JavaScript project.

Prompt

A prompt is any input you give to the agent. This includes direct questions ("What does this function do?"), instructions ("Refactor this to use async/await"), and more complex multi-step requests ("Build a REST endpoint for user authentication with input validation and error handling"). The quality of your prompts directly affects the quality of your agent's responses.

Session

A session is a single continuous interaction with your agent. It starts when you run openclaw and ends when you exit. During a session, the agent accumulates context from your conversation, the files it reads, and the actions it takes. Each session has its own context window.

Skill

A skill is a modular, reusable instruction set that teaches your agent a new capability. Skills are the building blocks of OpenClaw customization. Each skill is a markdown or TOML file containing structured instructions — system prompts, code examples, rules, and constraints.

Browse available skills at the OpenClaw Bazaar skills directory. Install a skill with openclaw skill install <skill-name>. Skills are open source, community-rated, and free.

Skill Categories

Skills are organized into categories in the Bazaar. Common categories include:

• Framework skills — Next.js, Django, Rails, Spring, etc.
• Language skills — TypeScript, Python, Rust, Go, etc.
• Testing skills — Jest, Pytest, Playwright, etc.
• DevOps skills — Docker, Terraform, CI/CD pipelines, etc.
• Security skills — OWASP checks, dependency auditing, etc.
• Style skills — Code formatting, naming conventions, review standards.

System Prompt

The system prompt is a set of instructions given to the agent before your conversation begins. It defines the agent's role, priorities, and constraints. Skills modify the system prompt by adding their instructions to it. When you install a skill, its contents become part of the system prompt your agent reads at the start of every session.

Tool

A tool is a specific capability the agent can use during a session. Built-in tools include reading files, writing files, running shell commands, and searching code. MCP servers add additional tools. When your agent says "I will use the file search tool," it is referencing one of these capabilities.

Workspace

A workspace is the root directory of the project where you run OpenClaw. The agent treats this directory as its scope — it reads files within the workspace, writes changes to it, and respects its configuration. Your workspace typically matches your Git repository root.

Putting It All Together

Here is how these terms relate in a typical workflow. You open a workspace and start a session. Your agent reads your project configuration, which includes installed skills and memory. You send a prompt. The agent uses its model and available tools (including any MCP servers) to respond within its context window. If you have hooks configured, they run at the appropriate points. If you chose a persona, all of this was pre-configured for you.

Understanding this vocabulary makes every other OpenClaw guide, tutorial, and discussion easier to follow. When you encounter a new term, check back here.

Browse the Skills Directory

Find the right skill for your workflow. The OpenClaw Bazaar skills directory has over 2,300 community-rated skills — searchable, sortable, and free to install.

Browse Skills →

Want a Pre-Built Setup?

If you would rather skip the browsing, OpenClaw personas come with curated skill sets already configured. Pick a persona that matches your role and start working immediately. Compare personas →