Copy this into your repo root to give coding agents clear project rules, commands, and red lines.
# AGENTS.md ## Project goal Describe what this repo does, who uses it, and what must never break. ## Stack - Language/runtime: - Framework: - Package manager: - Database/services: - Deployment target: ## Commands - Install: `...` - Test: `...` - Typecheck: `...` - Lint: `...` - Build: `...` ## Working rules for AI coding agents - Prefer the smallest safe change that solves the issue. - Read related files before editing. - Do not invent APIs, config names, or environment variables. - Do not delete migrations, production data, or auth/security code without explicit human approval. - Run the smallest meaningful verification before claiming success. - If tests cannot run, state why and give the exact command attempted. ## Red lines - Never commit secrets. - Never spend money or create paid cloud resources. - Never weaken authentication, authorization, payment, or privacy controls just to make tests pass. ## Definition of done - Requested change implemented. - Tests/checks run or clearly blocked. - Edge cases noted. - Files changed summarized.
Need the full pack with CLAUDE.md, Cursor rules, issue templates, safe command policy, and PR checklist? Get the €4.50 starter kit.