llmglot

A CLI tool for syncing canonical LLM agent configuration files across different coding assistants.

Why llmglot?

Managing configurations for multiple AI coding assistants is tedious. Each tool has its own location for instructions and skills:

• Claude uses CLAUDE.md and .claude/skills/
• Cursor uses .cursorrules and .cursor/rules/
• Windsurf uses .windsurfrules and .windsurf/rules/
• And so on...

llmglot lets you maintain a single source of truth (LLMGLOT.md + skills/) and sync it everywhere.

Installation

pip install llmglot

Or with uv:

uv add llmglot

Quick Start

1. Create your canonical config:

# Create LLMGLOT.md with your project instructions
echo "# My Project\n\nProject-specific instructions for AI assistants." > LLMGLOT.md

# Create skills directory
mkdir -p skills
echo "# Code Review\n\nWhen reviewing code, check for..." > skills/code-review.md

2. Sync to all agents:

llmglot sync all

That's it! Your config is now available to Claude, Cursor, Windsurf, Copilot, Cline, and Gemini.

Usage

# Preview what will be synced (recommended first step)
llmglot sync all --dry-run

# Sync everything to all agents
llmglot sync all

# Sync only to specific agents
llmglot sync all --target claude --target cursor

# Sync only skills or only the main instructions
llmglot sync skills
llmglot sync agents

Supported Agents

Templating

Use Jinja2 templates to customize content per agent:

# Project Guide

{% if agent_name %}
You are **{{ agent_name }}**.
{% endif %}

{% if agent_id == "claude" %}
Use /help for assistance.
{% elif agent_id == "cursor" %}
Use Cmd+K for inline edits.
{% endif %}

Available Variables

Project Structure

your-project/
├── LLMGLOT.md          # Canonical instructions (synced to all agents)
├── skills/             # Reusable skills
│   ├── code-review.md
│   └── testing.md
├── CLAUDE.md           # Generated - don't edit directly
├── .cursorrules        # Generated - don't edit directly
└── ...

License

MIT