> ## Documentation Index
> Fetch the complete documentation index at: https://docs.coreflux.org/llms.txt
> Use this file to discover all available pages before exploring further.

# Coreflux Agent Skills

> Install the Coreflux agent skills so coding agents write valid LoT and work with the broker

## Teach Your Agent LoT

Coding agents are good at general-purpose languages but have never seen LoT (Language of Things). The [documentation MCP](/ai/mcp) lets an assistant look syntax up while it works. **Agent skills** go one step further: they put the LoT and broker knowledge in the agent's own skills folder, so the agent loads it whenever a task calls for it.

[coreflux-skills](https://github.com/CorefluxCommunity/coreflux-skills) is the central, public repository of Coreflux [Agent Skills](https://agentskills.io). Its skills teach Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI, and other agents to write LoT and to work with the Coreflux MQTT broker. It is MIT licensed.

> [!NOTE]
>
> The Coreflux agent skills are available for broker **2.22 or later**.

> [!TIP]
>
> **Like handing a new engineer the plant's field manual on day one.** They still ask questions (the MCP), but they already know the vocabulary, the rules, and the safe way to make a change.

### When to Use This

* You write LoT with a coding agent and want it to **get the syntax right the first time**
* You want the agent to **validate before it deploys** LoT through the broker's [MCP Server](/latest/mqtt-broker/mcp-server)
* You want **one source of truth** for Coreflux skills across Claude Code, Cursor, VS Code, and other agents
* You need skills that **match a specific broker release**

***

Pick the skills your work needs, then install them with the method that fits your agent.

## Available Skills

The repository holds one folder per skill under `skills/`. `skills.json` lists them all.

| Skill | Use it for |
| - | - |
| `lot-actions` | Actions: triggers, variables, topic operations, control flow, Python and LLM calls |
| `lot-models` | Models: data schemas, field types, topic bindings, storage, inheritance |
| `lot-routes` | Connectors: databases, OT (Modbus, S7, OPC UA, …), REST, MQTT bridges, AI |
| `lot-rules` | Permission rules: ALLOW/DENY, topic patterns, priorities |
| `lot-agents` | `DEFINE AGENT`: LLM connectors, tool grants, delegation |
| `lot-things` | `DEFINE THING`: templates, parameters, references |
| `lot-panels` | LoTV panels: components, bindings, events |
| `lot-themes` | LoTV themes: design tokens, light and dark modes, variants |
| `python-scripts` | Python scripts the broker runs, and `CALL PYTHON` |
| `broker-tools` | The broker's MCP tools and the safe validate → deploy workflow |
| `broker-operations` | Operating a broker over MQTT: `$SYS/Coreflux/Command` and its replies, the `$SYS/Coreflux` topic map, triage, users and permissions, projects and `.lotnb` notebooks |

## Install the Skills

**Any agent**

Use the skills CLI. It works with any agent that reads Agent Skills:

```bash wrap theme={"theme":"css-variables","languages":{"custom":["/languages/lot.json"]}}
npx skills add CorefluxCommunity/coreflux-skills
```

**Claude Code**

Add the repository as a plugin marketplace, then install the `coreflux` plugin:

```text wrap theme={"theme":"css-variables","languages":{"custom":["/languages/lot.json"]}}
/plugin marketplace add CorefluxCommunity/coreflux-skills
/plugin install coreflux@coreflux
```

**VS Code and Cursor**

The [LoT Notebooks extension](/lot-notebooks/usage#develop-lot-notebooks-with-ai) has an **Install Agent Skills...** button in its sidebar. Pick the skills to install and the target folders: the `.agents`, `.claude`, `.cursor`, or `.github` skills folder, in the workspace or in your home folder.

**By hand**

Copy `skills/<name>/` from the repository into your agent's skills folder: `.agents/skills/`, `.claude/skills/`, or `.cursor/skills/`. Use the folder in your project for one project, or the one in your home directory for all of them.

### Match Your Broker Version

Tags in the repository match broker releases: the tag for a release has the same name as its broker version. To match the broker you run, add that version after `#`. For example, for broker 2.22.0:

```bash wrap theme={"theme":"css-variables","languages":{"custom":["/languages/lot.json"]}}
npx skills add CorefluxCommunity/coreflux-skills#2.22.0
```

Without a tag, you get the default branch, which follows the broker's main branch.

## Where Skills Come From

The skills are generated from the Coreflux broker. The broker ships the same skills to its built-in agents, and only the skills meant for external agents are exported to coreflux-skills. The export runs on every change to the broker's main branch and on every broker release tag.

> [!WARNING]
>
> Do not open pull requests against coreflux-skills. The next sync overwrites them. Report a problem as an [issue](https://github.com/CorefluxCommunity/coreflux-skills/issues) instead.

A client connected to the broker's [MCP Server](/latest/mqtt-broker/mcp-server#what-the-tools-do) can also read the broker's skill catalogue from the `coreflux://skills` resource.

***

## Best Practices

> **Pair skills with the MCP**
>
> Skills give the agent the rules up front. The [documentation MCP](/ai/mcp) answers the questions the skills don't cover. Use both.

> **Install only what you use**
>
> An agent that writes Actions and Models does not need the panel and theme skills. Fewer skills keep the agent's context focused.

> **Keep project rules separate**
>
> Skills describe LoT and the broker. Your naming, topic layout, and boundaries belong in an [AGENTS.md](/ai/best-practices#setting-up-your-ai-assistant) file.

> **Pin a tag for production**
>
> When the agent works against a production broker, install from the tag that matches its version so the skills never describe syntax that broker does not have.

***

## Next Steps

> **[Coreflux MCP](/ai/mcp)**
>
> Connect your assistant to the Coreflux documentation for real-time lookups.

> **[MCP Server](/latest/mqtt-broker/mcp-server)**
>
> Let an AI client validate and deploy LoT on a live broker.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.