Quick Start

Prerequisites

• Node.js >= 18

Installation

npm install @johannes.latzel/llm-chat-skill

Basic usage — plain skill

Create a skill file at ./skills/my-tool/SKILL.md:

---
name: my_tool
description: A demonstration skill
---

This skill provides guidance on using the my_tool tool suite.

Then load it programmatically:

import { SkillRegistry, SkillRegistryConfiguration, LoadSkillTool } from '@johannes.latzel/llm-chat-skill';

const config = new SkillRegistryConfiguration('./skills');
const registry = new SkillRegistry(config);
await registry.initialize();

const tool = new LoadSkillTool(registry);
const result = await tool.execute({ name: 'my_tool' });
console.log(result.result);
// "This skill provides guidance on using the my_tool tool suite."

Structured skill

A structured skill breaks its body into separate sections managed via the resource tools. Create one via set_skill by omitting the body:

import { SkillRegistry, SkillToolPackage, SetSkillTool, SetSkillResourceTool, LoadSkillTool } from '@johannes.latzel/llm-chat-skill';

const registry = new SkillRegistry();
await registry.initialize();

// Create a structured shell (no body)
const setTool = new SetSkillTool(registry);
await setTool.execute({
    name: 'code_review',
    description: 'Code review guidelines'
});

// Add sections
const resourceTool = new SetSkillResourceTool(registry);
await resourceTool.execute({
    skill_name: 'code_review',
    resourceType: 'sections',
    name: 'purpose.md',
    content: '# Purpose\n\nEnsure code quality.'
});
await resourceTool.execute({
    skill_name: 'code_review',
    resourceType: 'sections',
    name: 'rules.md',
    content: '# Rules\n\n1. No debug logs\n2. Handle errors'
});

// Load it — body is auto-composed from sections
const loadTool = new LoadSkillTool(registry);
const result = await loadTool.execute({ name: 'code_review' });
console.log(result.result);
// "## Purpose\n\nEnsure code quality.\n\n## Rules\n\n1. No debug logs\n2. Handle errors\n\n--- Resources ---\n\nsections/purpose.md\nsections/rules.md"

Tutorial

The SkillToolPackage provides a tutorial via tutorial() that explains how to use the skill system tools:

import { SkillRegistry, SkillToolPackage } from '@johannes.latzel/llm-chat-skill';

const registry = new SkillRegistry();
const pkg = new SkillToolPackage(registry);

console.log(pkg.tutorial()); // Full guide content

Using the SkillToolPackage

For integration with a ChatService from @johannes.latzel/llm-chat, use the SkillToolPackage to register all six tools at once:

import { SkillRegistry, SkillToolPackage } from '@johannes.latzel/llm-chat-skill';

const registry = new SkillRegistry(config);
await registry.initialize(service);       // loads skills + wires "skills" prompt
service.tools().add(new SkillToolPackage(registry));  // registers all 6 tools

Passing your ChatService to initialize() adds a skills prompt to the system prompt with the current skill listing, and automatically refreshes it when skills are created, updated, or deleted via tool calls.

If you are using ToolSuite directly (without a ChatService):

import { ToolSuite } from '@johannes.latzel/llm-chat';
import { SkillRegistry, SkillToolPackage } from '@johannes.latzel/llm-chat-skill';

const registry = new SkillRegistry();
await registry.initialize();

const suite = new ToolSuite();
suite.add(new SkillToolPackage(registry));

// suite now has all 6 skill tools registered
const tools = suite.getTools(); // OpenAI-compatible tool definitions

Reading reference files

If your skill has reference documents:

import { GetSkillResourceTool } from '@johannes.latzel/llm-chat-skill';

const readTool = new GetSkillResourceTool(registry);
const guide = await readTool.execute({ skill_name: 'my_tool', resourceType: 'references', name: 'guide.md' });