AI Agent Skill Library

Curated skills for AI coding agents: architecture patterns (Clean, Hexagonal, CQRS, Event-driven) and testing strategies (Unit, Integration, E2E).

Each skill is a reusable prompt fragment that teaches an agent a specific practice.

Quick Start

uv sync # install dependencies
make install-mcp # install & deploy MCP servers
make install-skills # install skills
make install # install both

Make Targets

MCP Servers

MCP (Model Context Protocol) servers are configured in mcp/mcp.json. On install, each server's runtime is verified (npx, uvx, or custom command) and its config is deployed to every supported agent platform.

Supported agent platforms

Platform-scoped servers

Add a "platform" field to restrict a server to a single agent:

{
 "mcpServers": {
 "my-opencode-tool": {
 "command": "npx",
 "args": ["-y", "some-package"],
 "platform": "opencode"
 }
 }
}

Servers without a platform are shared — deployed to every agent. Servers with a platform only go to that agent's config.

See docs/mcp-servers.md for full details on the config format, server fields, multi-agent deployment, and adding new agent platforms (guide).

Configuration

Place secrets in .env:

GITHUB_TOKEN=...
FORGEJO_ACCESS_TOKEN=...

The mcp/mcp.json template uses Jinja2 expressions — missing variables fall back to human-readable placeholders.

Documentation