The AGENTS.md Playbook — make your AI coding agent actually reliable

Stop re-explaining your codebase to your AI every session.

Write one context file your agent actually follows — and it stops inventing a second HTTP client, running the wrong test command, and editing the generated file. 8 stack templates get you there in 30 minutes.

Get the Playbook Pay what you want · from $12+
AGENTS.md▏ 
# Commands
install: pnpm install      ← never npm
test:    pnpm test --run

# Conventions
- Business logic lives in services/,
  never in route handlers
- Reuse lib/http.ts — do not add axios

# Do NOT touch
- *.generated.ts
- migrations/  (ask first)

Same agent. One file of difference.

Without a context file
  • Invents a second HTTP client next to the one you already have
  • Runs npm test when the repo is pnpm
  • Edits a *.generated.ts file by hand
  • You correct it — then repeat tomorrow, from zero
With AGENTS.md
  • Reuses lib/http.ts like a teammate would
  • Runs pnpm test --run the first time
  • Leaves generated files and migrations alone
  • Behaves the same way on every new session

The method

The anatomy of a context file your agent won't ignore.

Not a wish list — a structure tested against real agent behavior. Five parts, in order:

01

Orientation

What this repo is, which dir is which.

02

Commands

The exact build/test/run lines.

03

Conventions + exemplars

Rules paired with a file to copy.

04

Guardrails

The do-not list that prevents real damage.

05

Glossary

Your project's words, defined.

8 fill-in-the-blank templates

For the stacks you actually ship. Copy the closest one, fill the blanks, done.

▸
Generic
▸
Node / TypeScript API
▸
Next.js full-stack
▸
Python service
▸
Monorepo root
▸
Monorepo package
▸
Data pipeline
▸
Infrastructure / IaC
Claude Code Cursor Copilot Aider Windsurf

Or let your agent write it for you

The kit includes an AI helper file: paste it into Claude Code or Cursor, and it inspects your repo and drafts your AGENTS.md — you just verify. Plus a robust FAQ for monorepos, secrets, memory vs skills, and "why does it still ignore my rules?"

Pay what you want

Minimum $12. Suggested $29 — less than one wasted afternoon. Instant download · free updates.

$12floor $19fair $29suggested

Leave it blank to decide your price on the next step.

Just take me to checkout

Secure checkout by Gumroad · PDF + 8 templates + AI helper + FAQ (zipped)

Questions

Which filename should I use — AGENTS.md, CLAUDE.md, .cursorrules?

Whichever your tool auto-loads — the Playbook maps the common ones and the FAQ has a decision guide. The content structure is identical across them; only the filename changes. Many people keep one canonical file and symlink or mirror it.

My agent still ignores the rules. Why?

Usually the file is a vague wish list. The Playbook's fix is rules paired with exemplars (point at a real file to copy) and a tight guardrail list — that's what actually changes behavior. There's a whole section on debugging a file against live agent output.

Does it work for monorepos?

Yes — there are dedicated root and per-package templates, plus guidance on what belongs where so you don't repeat yourself across packages.

Is this tied to one AI tool?

No. It covers any agent that reads a project instruction file — Claude Code, Cursor, Copilot, Aider, Windsurf. The concept (a file your agent auto-loads) is stable even as filenames evolve.

Write it once. Stop repeating yourself.

Thirty minutes now buys you an agent that behaves on every session from here on.

Get the Playbook — pay what you want