Give Your AI Context

Generic answers come from context-starved AI. The fix is a small file in your repo — AGENTS.md or CLAUDE.md — that works like a README for the AI: your stack, your conventions, your commands, stated once instead of retyped into every chat.

What goes in it

  • Tech stack and key conventions — how you name things, where files go
  • Commands to run before committing (lint, tests)
  • Links to ND skills and your architecture docs
  • What not to do — the mistakes you keep seeing

A real example

# Donor Importer

Stack: Rails 7, Oracle, RSpec
Auth: ND SSO via Okta

Conventions:
- Service objects in app/services
- Specs mirror app/ structure
- bin/lint && bin/test before commit

See: skills/oit-architecture.md
See: skills/nd-user-experience.md

The test for what belongs: if a new developer would need to know it, the AI needs to know it. Keep it current the way you'd keep a README current — when a convention changes, update this file.

Why this works

Agentic tools read these files automatically at the start of a session, so every chat begins already knowing your project instead of guessing at a generic one. Combined with one chat per feature, it means each conversation starts smart and stays focused.

The Toolkit has a starter template and ND-published skills you can reference from your context file so the AI knows how Notre Dame builds software — not just how software gets built.

Ten minutes, one repo

  1. Pick a repo you own and create AGENTS.md at its root.
  2. Write the five things you'd tell a new teammate on day one.
  3. Add the pre-commit commands and one “common mistake” you're tired of correcting.
  4. Next AI session, notice how much less explaining you do.