THE AI LAB : TOOLS, SKILLS, RECIPES FOR AI AGENTS

Project's Goals

This repository is a laboratory to experiment with tools, skills, and templates, to enhance our experience of working with AI Agents (such as OpenCode, Claude Code, etc.). Our goal is to identify, test, and configure some optimal tools for specific tasks like coding, web development, research, and daily organization.

What's inside

• recipes/ — deployable bundles of agents, skills, and commands. One directory per recipe, ready to copy into any project without renaming.
• tools/ — source-code tool projects used by the lab and by recipes (MCP servers).
• docs/ — indexed tool documentation, maintained by The Librarian.
• AGENTS.md — detailed instructions and personas for the agents used in this lab.

🧩 Recipes

Each recipe has its own README that states its intent, usage examples, and components.

• The Marketing Studio: Multi-agent pipeline for brand-aligned marketing visuals.
• The Designer: Local, autonomous design systems and style guides.
• The Librarian: Specialized sub-agent for local documentation indexing and management.
• Git Commit Workflow: Atomic, conventional commits with mandatory human-in-the-loop confirmation.
• Agent Browser: Browser automation CLI for AI agents.
• Technical Writing: Controlled technical English based on ASD-STE100 Simplified Technical English.
• AI Lab: Create, install, test, remove, and improve recipes with the /lab command.
• Vercel: Vercel platform specialist. Task-oriented skills for the CLI, the AI SDK, and Serverless Functions.

To deploy a recipe into your project you have two main options:

• deploy the recipe manually or tell your agent to follow the instructions : Deploying a Recipe.
• install the /lab command to fully manage your recipes (install, test, remove, create, improve) :

Deploying a Recipe

Deploy a recipe into a target project. The target project is where the agent, skills, or commands run. Run every deploy command from the root of the target project. The recipe source lives in this lab under recipes/<name>/.

The preferred way is the /lab command. It works in any project that has the @the-ai-lab reference and the /lab command installed (see Install the /lab command). The command routes each action to the correct skill. It loads the deploy-recipes skill to install, test, or remove a recipe, and the manage-recipes skill to create a new one or to improve an existing one. The deploy-recipes skill ships reusable scripts that perform every file operation. Use /lab install <recipe-name> to deploy, /lab test <recipe-name> to symlink for testing, /lab remove <recipe-name> to remove a recipe, /lab create <recipe-name> to create a new recipe in the lab, and /lab improve <recipe-name> to add reusable scripts to a recipe. The methods below describe the underlying mechanics that the command automates.

Component	Source (this lab)	Destination (target project)
Agent	recipes/<name>/agents/*.md	.opencode/agents/
Skill	recipes/<name>/skills/<skill>/	.agents/skills/<skill>/
Command	recipes/<name>/commands/*.md	.opencode/commands/

The recipe pre-names every file for its destination. A direct copy needs no renaming.

Method 1: Deploy from the Git repository (recommended)

This method needs no local copy of the lab. OpenCode exposes the lab to the target project as a project reference.

1. Add the-ai-lab as a project reference in the target project's opencode.json. Create the file if it does not exist.

   {
     "references": {
       "the-ai-lab": {
         "repository": "zipang/the-ai-lab",
         "description": "Use to deploy recipes (agents, skills, and commands) from The AI Lab"
       }
     }
   }

The description tells the agent when to use the reference. OpenCode clones the repository into its cache and makes it available as @the-ai-lab. To make @the-ai-lab available in every project, add the same reference to the global config at ~/.config/opencode/opencode.json instead of the project file.

2. At any moment you can now ask your agent to deploy a recipe from the @the-ai-lab reference. Example:

   Install the technical-writing recipe from @the-ai-lab into this project.
   

   If the /lab command is installed, use /lab install technical-writing instead.

Method 2: Deploy from a local copy of the lab

Use this method when you already have a checkout of the lab on disk.

Example: deploy the technical-writing recipe.

# Run from the root of the target project.
mkdir -p .agents/skills
cp -r <path-to-lab>/recipes/technical-writing/skills/* .agents/skills/

Install the /lab command

Copy and paste this prompt to let your agent install the /lab command in your project:

Install the /lab command from @the-ai-lab into this project.

1. Add `the-ai-lab` as a reference in `opencode.json`. Create the file if it does not exist.
   Use this configuration:

   {
     "references": {
       "the-ai-lab": {
         "repository": "zipang/the-ai-lab",
         "description": "Use to deploy recipes (agents, skills, and commands) from The AI Lab"
       }
     }
   }

   The `description` is required. It tells the agent when to use the reference
   and gives the agent access to the resolved reference path.

2. Copy the command from `@the-ai-lab/recipes/lab/commands/lab.md` into
   `.opencode/commands/lab.md`. If `@the-ai-lab` is not available yet, fetch the
   file from https://raw.githubusercontent.com/zipang/the-ai-lab/master/recipes/lab/commands/lab.md
   and write it to that path.

3. Copy the skills from `@the-ai-lab/recipes/lab/skills/deploy-recipes/` and
   `@the-ai-lab/recipes/lab/skills/manage-recipes/` into `.agents/skills/`.
   The command routes each action to one of these skills to perform its work.

4. Report how to use the command:
   /lab install <recipe-name>
   /lab test <recipe-name>
   /lab remove <recipe-name>
   /lab create <recipe-name>
   /lab improve <recipe-name>

🛠️ Local MCP Servers

Tools that require a local installation and build step to be installed. See tools/ for details.

• Bun & TypeScript Template: Quick-start template for building local MCP servers with Bun.
• Rust Filesystem MCP: A high-performance Rust implementation for filesystem operations.
• MCP Inspector: Interactive debugger for testing and exploring MCP servers.
• Image Studio: Multi-provider AI image generation with cost tracking.