Stack.menu has launched a new resource designed to bridge the gap between human architectural intent and AI coding agent execution. The platform offers curated stack guides that act as ready-made briefs for agents like ChatGPT, Claude, Cursor, and Codex. By providing structured guidance, it aims to prevent agents from defaulting to generic solutions when a specific tech stack is required for a project's success.

The llms.txt Standard for Agent Context

The core utility of Stack.menu is its implementation of the llms.txt standard, specifically available at https://stack.menu/llms.txt. This file lists required stack guides and instructs agents on how to select, build with, or plan rewrites using them. Developers can paste this link directly into their agent of choice, ensuring the AI has immediate access to curated, readable stack documentation rather than relying on its potentially outdated or generic training data.

Curated Stacks and Real-World Application

The site features a library of 14 curated stacks, covering tools, setup instructions, and validation checks. A demonstration video on the site shows a real ChatGPT session where the agent uses Stack Menu to recommend Phoenix + LiveView with PostgreSQL for a family meal planner. The agent explicitly compares this option against Next.js and Rails + Hotwire, taking the user's existing Elixir experience into account, and notes that offline shopping capabilities would require additional client support.

Key Takeaways

  • Stack.menu provides 14 curated stack guides to help coding agents choose appropriate technologies.
  • The llms.txt file allows developers to point agents like Claude, ChatGPT, and Cursor to specific stack requirements.
  • Agents can use these guides to compare options (e.g., Phoenix vs. Next.js) based on user experience and project needs.
  • The resource is designed to be agnostic, supporting multiple major AI coding tools.

The Bottom Line

This is a smart play for the emerging 'agent context' layer; instead of fighting the AI's hallucinations, Stack.menu gives them a map, making it easier to enforce architectural consistency without writing massive system prompts.