Creating Modules
A Lola module can be an Agent Plugins 1.0 package, a single Agent Skill, or a legacy AI Context Module.
An Agent Plugins package is the default lola mod init layout: a
plugin.json manifest with portable skills/ and mcp.json at the root,
plus Lola's commands, agents, and instructions under the dev.getlola
extension namespace. An Agent Skill is a standalone SKILL.md with optional
supporting files. An AI Context Module is Lola's legacy layout - it wraps one
or more skills alongside AGENTS.md, commands/, and mcps.json inside a
module/ directory, and now requires lola mod init --format lola.
Initialize a Module
By default this emits an Agent Plugins 1.0 package:
my-plugin/
plugin.json
skills/
example-skill/
SKILL.md
mcp.json
dev.getlola/
commands/
example-command.md
agents/
example-agent.md
AGENTS.md
The name in plugin.json is the module identity used by Lola when the
package is registered. Portable skills and MCP servers stay at the package
root. Lola-specific commands, agents, and instructions live in Lola's
dev.getlola extension namespace.
Initialize a Legacy AI Context Module
This creates Lola's legacy AI Context Module structure:
my-module/
module/
AGENTS.md # AI main spec
skills/
example-skill/
SKILL.md # Skill following agentskills.io
commands/
example-command.md
agents/
example-agent.md
mcps.json # MCP settings
Edit the skill
Edit your skill's SKILL.md following the AgentSkills.io standard:
---
name: my-skill
description: When to use this skill
---
# My Skill
Instructions for the AI assistant...
Add supporting files
Each skill can have its own scripts/, reference/, and assets/ directories:
my-skill/
SKILL.md
scripts/ # Executable scripts
reference/ # Documentation
assets/ # Other supporting files
Reference them with relative paths in your SKILL.md:
Add an AGENTS.md
The AGENTS.md provides module-level context that applies across all skills in the module. This is what elevates a collection of skills into an AI Context Module.
Add slash commands
Slash commands are custom commands that can be invoked with /command-name in AI assistants. Claude Code, Cursor, and Gemini CLI all support them. Create markdown files in commands/:
---
description: Review a pull request
argument-hint: <pr-number>
---
Review PR #$ARGUMENTS and provide feedback.
Use $ARGUMENTS for all args or $1, $2 for positional. Lola automatically converts commands to each assistant's native format (markdown for Claude Code and Cursor, TOML for Gemini CLI).
Register and install
See Skill Format for details on the SKILL.md specification.