agloom

Warning: This project is in active development. Its API may introduce breaking changes between minor versions. Use at your own risk.

Transpile canonical agent configurations across AI coding assistants.

Write your instructions, skills, agents, MCP servers, and permissions once in a single .agloom/ directory — agloom generates the correct config files for each target tool.

The problem

AI coding assistants each expect their own config format: Claude Code reads CLAUDE.md and .claude/, OpenCode reads AGENTS.md and .opencode/, Gemini CLI reads GEMINI.md and .gemini/, KiloCode reads .kilo/, and so on. Maintaining these by hand is tedious and error-prone.

The solution

agloom introduces a canonical format — Markdown and YAML files under .agloom/ — and transpiles them into agent-specific outputs. One source of truth, multiple outputs. Think Sass → CSS, or TypeScript → JavaScript.

Quick start

npm install -g agloom

mkdir -p .agloom/instructions
echo "adapters:\n  - claude" > .agloom/config.yml
echo "# My Project\n\nInstructions for AI assistants." > .agloom/instructions/AGLOOM.md

agloom transpile

This generates CLAUDE.md and .claude/ from your canonical config. See the Getting Started guide for a full walkthrough.

Supported adapters

Adapter	Description	Key outputs
claude	Claude Code	CLAUDE.md, .claude/
opencode	OpenCode	AGENTS.md, .opencode/
gemini	Gemini CLI	GEMINI.md, .gemini/
kilocode	KiloCode	AGENTS.md, .kilo/
codex	Codex CLI	AGENTS.md, .codex/
agentsmd	Any AGENTS.md-compatible tool	AGENTS.md

Transpilers

agloom processes six types of canonical config:

Transpiler	Source	Description
Instructions	.agloom/instructions/	System instructions per agent
Skills	.agloom/skills/	Reusable slash commands
Agents	.agloom/agents/	Sub-agent definitions
Commands	.agloom/commands/	Custom CLI commands
MCP	.agloom/mcp.yml	MCP server configuration
Permissions	.agloom/permissions.yml	Tool permission rules

Documentation

Full documentation is available at docs.agloom.sh.

Guide

Step-by-step tutorials for learning agloom:

• Introduction — The problem, the solution, and core principles
• Getting Started — From zero to first transpile in 5 minutes
• Project Structure — Anatomy of the .agloom/ directory
• Instructions — Writing instructions with agent-specific blocks
• Skills & Agents — Creating reusable skills and sub-agents
• Plugins — Using and creating plugins
• Overlays — Per-adapter customization with merge, override, and patch
• Variables — Interpolation and the values system

Reference

Complete specifications for every feature:

• CLI Commands — All commands, flags, and exit codes
• Configuration — Full .agloom/config.yml schema
• Plugin Manifest — Full plugin.yml schema
• Adapters — Adapter capabilities and output paths
• Interpolation — Variable namespaces, syntax, and resolution
• Patch Operations — $set, $merge, $append, and more
• Transpilers — Pipeline modules and processing order

License

Apache 2.0