AI

AI Overview

Use GoLazy's agent guidance, default skills, authentication, and MCP server support without hiding application ownership.

By Guillermo Alvarez - Published - Updated

AI features in GoLazy are application-facing interfaces, not a separate runtime that takes ownership away from the app. The application still owns its source code, data access, authentication policy, authorization decisions, and deployment boundaries.

Use this section when a GoLazy app needs to be understandable to coding agents, serve tools to AI clients, or prepare for AI-assisted workflows.

Generated app guidance

New apps include AGENTS.md guidance at the application root. That file is the shared instruction surface for coding agents and automation. It documents the app shape, route and controller conventions, asset commands, service task shape, and verification expectations.

Generated apps also keep CLAUDE.md as a thin tool-specific pointer back to AGENTS.md. Durable guidance that applies to every agent belongs in AGENTS.md; tool-specific notes should stay small.

Default skills

New apps include a default GoLazy framework skill at .skills/golazy-framework/SKILL.md. The skill is for repeated coding-agent work such as adding routes, controllers, services, views, assets, and tests.

Keep AGENTS.md concise. When a workflow needs long references, scripts, templates, or examples, move that workflow into one .skills/<name>/SKILL.md entrypoint and keep supporting files beside it.

Authentication and MCP

Use Authentication And OAuth when an AI client or dependent application needs to sign in. lazyauth authenticates identities, lazyoauth serves authorization-code and PKCE flows, and lazyjwt validates tokens and puts claims in context.

Use MCP Server when the app should expose tools, resources, prompts, skills, or MCP Apps UI resources through /mcp. GoLazy only mounts the MCP middleware when the app registers at least one MCP module.

Future response generators

Response generators are not part of the shipped AI guide path yet. Until they exist, keep model calls, prompt construction, retrieval, and response shaping inside explicit application services or MCP tools. When the generator package lands, this section should gain a focused guide that shows how generated responses compose with services, tests, and MCP exports.