Write useful AGENTS.md instructions
AGENTS.md records reusable project guidance. Include verified commands, conventions, boundaries, and completion criteria. It is guidance, not a security barrier.
What is it?
AGENTS.md records reusable project guidance. Include verified commands, conventions, boundaries, and completion criteria. It is guidance, not a security barrier.
Why it matters
Reusable rules help keep work consistent across sessions.
When to use it
When the same rules apply to repeated project tasks.
How to do it
- Read the explanation and identify the files or environment involved.
- Use the generator for your project type. Replace every placeholder, inspect the real configuration for available commands, and save a reviewed AGENTS.md at the project root.
- Compare the result with the verification criteria below. Record what worked and what remains unclear.
Worked example
Use the generator for your project type. Replace every placeholder, inspect the real configuration for available commands, and save a reviewed AGENTS.md at the project root.
Try it yourself
Repeat the exercise with your own wording or a second sample. Use the smallest change that demonstrates the idea.
Prompt builderVerify your result
A new task can list the instruction sources it loaded. Each documented command exists in the project.
Common mistakes
Copying npm commands into a non-Node project. Use commands verified for the actual stack.
How to recover
Stop at the first unexpected result. Preserve your current work, re-check the named file or environment, and repeat only the failing step. Ask for an explanation if the recovery is unclear.
Official references
Product interfaces and access can change. Use the linked official guide for current installation and account requirements.
OpenAI · AGENTS.md ↗Source review: 2026-09-16