Guides
Documentation
Customizing agents

Agent instructions

What Fragua writes into a new app so every agent run starts oriented.

When Fragua scaffolds a new app (the greenfield path), it commits one small file of agent instructions before the first push. It's there so every later run — yours or a teammate's — starts knowing the basics of the app instead of rediscovering them each time.

What gets created

Two files, committed together. One is the real document; the other is a pointer to it, so every coding agent finds the same instructions where it expects them.

.claude/CLAUDE.md

The instructions themselves. Claude Code reads this file automatically at the start of a run, with no setup on your part.

AGENTS.md

A symlink at the repo root pointing to the file above. Other coding agents look for AGENTS.md there, so they read the same instructions — one source, no duplicate to keep in sync.

What's inside

A deliberately short starting point — the app's name, how to run it, and room for you to grow it. Nothing you have to maintain on day one.

Commands
bin/dev Start the development server on the app's assigned port.
bin/rails server Start the server without the process manager.
bin/rails test Run the test suite.
bin/setup Install dependencies, prepare the database, and get a fresh checkout running.
  • The app's development port is fixed for you, so it never collides with another app you're running side by side.

  • An empty Gotchas section is left for you to fill in — the place to record surprises as you hit them. The next guide covers what's worth adding.

Why two files

Claude Code loads the .claude/CLAUDE.md file on its own, so the instructions reach the agent without any configuration. Other tools — like Codex or Cursor — look for an AGENTS.md at the repository root instead. The symlink bridges the two: both read the exact same content, and you only ever edit one file.

Make it your own

A few lines of project context measurably improve every run.

Tuning agent instructions