AGENTS.md

Next.js app (App Router). Replace the bracketed bits with your details.

Stack

• Next.js 14 (App Router), React 18, TypeScript strict.
• Package manager: pnpm 9. Tests: Vitest (unit) + Playwright (e2e).
• Lint/format: ESLint (next config) + Prettier.

Project Structure

• app/ — App Router routes, layouts, and route handlers (server by default).
• components/ — shared React components; co-locate route-specific ones under app/.
• lib/ — server utilities, data access, and NEXT_PUBLIC_*-aware config.
• e2e/ — Playwright specs; unit tests live as *.test.tsx beside source.
• next.config.js, package.json — build config and scripts.

Setup

pnpm install
cp .env.example .env.local   # Next reads .env.local; never commit it

Commands

pnpm dev               # next dev (http://localhost:3000)
pnpm build             # next build (also type-checks)
pnpm start             # serve the production build
pnpm test              # vitest run
pnpm test:e2e          # playwright test
pnpm lint              # next lint
pnpm typecheck         # tsc --noEmit

Code style

• Default to Server Components; add "use client" only when you need state, effects, or browser APIs.
• Fetch data in server components / route handlers; never expose secrets to the client bundle (only NEXT_PUBLIC_* vars reach the browser).
• Co-locate components with their route segment under app/.

Example — a server component fetching data:

// app/orders/page.tsx (Server Component by default)
export default async function OrdersPage() {
  const orders = await db.order.findMany({ take: 20 }); // runs on the server
  return <OrderList orders={orders} />;
}

Testing

• Unit tests as *.test.tsx beside components; e2e specs under e2e/.
• A change is done when pnpm lint, pnpm typecheck, and pnpm test pass; run pnpm test:e2e for routing/flow changes.
• Add a failing test first, then make it pass (TDD).

Git & PRs

1. Branch from main: git switch -c feat/<short-name>.
2. Conventional Commits (feat:, fix:, chore:).
3. Before pushing: pnpm lint && pnpm typecheck && pnpm test.
4. PR description: one-line summary + the test command you ran.

Boundaries

• Always: edit app/**, components/**, and lib/**, and keep browser-safe config in NEXT_PUBLIC_* vars.
• Always: read .env.example and add new keys there with a placeholder value.
• Always: fix the underlying issue when ESLint or tsc complains.
• Ask first: before changing next.config.js, middleware, auth, or DB schema.
• Ask first: before adding a new dependency or bumping a major version.
• Never: reference a server-only secret inside a "use client" component.
• Never: disable ESLint rules inline just to ship.
• Never: commit secrets (.env.local and real keys stay out of git).

More

• Routing & data conventions: docs/architecture.md.

agents-md-cookbook

The tested, tool-agnostic AGENTS.md kit.

Copy a battle-tested AGENTS.md into your repo, lint it in CI, and migrate your old CLAUDE.md / .cursorrules in one command. Every template is short on purpose, leads with runnable commands, and is checked by our own linter.

AGENTS.md is a single Markdown file at your repo root that tells coding agents how to build, test, and behave in your project. It is stewarded by the Linux Foundation's Agentic AI Foundation, lives in 60,000+ repositories, and is read natively by 24+ tools. One file, every agent.

---

30-second value prop

• Tool-agnostic. One AGENTS.md works across Cursor, Codex, Copilot, Windsurf, Cline, Zed, Jules, Amp, and more. See the full compatibility matrix.
• Tested, not vibes. Templates follow evidence from GitHub's 2,500-repo study, ETH Zurich's AGENTbench, and Augment's golden-PR eval — and CI lints every template with our own agents-md-lint.
• Short by design. 60-150 lines, commands first, real code snippets, explicit boundaries. No platitudes, no architecture essays, no token bloat.
• Migrate in one command. Turn an existing CLAUDE.md, .cursorrules, or GEMINI.md into a clean AGENTS.md with agents-md-migrate.

Quick start

1. Copy a template for your stack into your repo root as AGENTS.md (filename must be exactly AGENTS.md — uppercase AGENTS, lowercase .md):

   curl -fsSL \
     https://raw.githubusercontent.com/Taiizor/agents-md-cookbook/main/templates/typescript-node/AGENTS.md \
     -o AGENTS.md
   

   Browse the full template index for other stacks.

2. Already have agent instructions? Migrate them instead of starting over:

   bunx agents-md-migrate          # or: npx agents-md-migrate
   

   This detects CLAUDE.md, .cursorrules, GEMINI.md, and friends, then produces a single AGENTS.md. (Published by the agents-md-migrate package.)

3. Lint it in CI so the file stays accurate as your project grows:

   # .github/workflows/ci.yml
   - name: Lint AGENTS.md
     run: bunx agents-md-lint AGENTS.md   # or: npx agents-md-lint AGENTS.md
   

   Prefer a one-line GitHub Action? Use the bundled action.yml:

   # .github/workflows/ci.yml
   - uses: Taiizor/agents-md-cookbook@v1
     with:
       path: AGENTS.md
   

Compatibility at a glance

Tool	Reads AGENTS.md?
Cursor	NATIVE
OpenAI Codex	NATIVE
GitHub Copilot (coding agent)	NATIVE
Windsurf / Cascade	NATIVE
Cline	NATIVE
Zed	NATIVE (first-match)
Amp	NATIVE
Google Jules	NATIVE
opencode / RooCode	NATIVE
Claude Code	ADAPTER (symlink / @AGENTS.md)
Aider	ADAPTER (--read AGENTS.md)
Gemini CLI	CONFIG (opt-in)
Copilot (VS Code chat)	CONFIG (experimental)

Full mechanism, nesting behavior, own-file fallbacks, and sources are in COMPATIBILITY.md (with a dated last-verified line).

Templates

Each template is a complete, copy-pasteable AGENTS.md for one stack.

Stack	Path
TypeScript + Node	templates/typescript-node/AGENTS.md
Python	templates/python/AGENTS.md
Go	templates/go/AGENTS.md
Rust	templates/rust/AGENTS.md
Java + Spring Boot	templates/java-spring/AGENTS.md
.NET / C#	templates/dotnet-csharp/AGENTS.md
Next.js	templates/nextjs/AGENTS.md
React + Vite	templates/react-vite/AGENTS.md
Django	templates/django/AGENTS.md
FastAPI	templates/fastapi/AGENTS.md
Ruby on Rails	templates/rails/AGENTS.md
Monorepo (root)	templates/monorepo/AGENTS.md
Data / ML	templates/data-ml/AGENTS.md
React Native / Expo	templates/react-native/AGENTS.md
Minimal starter	templates/minimal/AGENTS.md

Why AGENTS.md

Coding agents auto-execute the commands you list, so a single file that is accurate, short, and command-first beats scattered per-tool config. The research is blunt about what works:

• Lead with commands and flags. GitHub's review of 2,500+ repos found the best files put build/test commands early with exact flags, prefer code examples over prose, and pair every boundary with a "do."
• Keep it short and non-inferable. ETH Zurich's AGENTbench found auto-generated /init files reduce task success ~3% and raise cost ~20-23%; agents follow files faithfully, so every wrong line is executed. Limit content to what an agent cannot infer (custom tooling, unusual builds).
• Aim for 100-150 lines. Augment's golden-PR eval found a 100-150 line sweet spot, numbered workflows (+25% correctness), decision tables, and 3-10 line real-code examples (+20% reuse); gains reverse beyond ~300 lines.

Read the full handbook: anatomy · best practices · nesting & monorepos · common mistakes.

Tooling

Published to npm so you can run them with bunx or npx (no install needed):

• agents-md-lint — scores an AGENTS.md against the evidence-based rules and fails CI on regressions.
• agents-md-migrate — converts existing CLAUDE.md / .cursorrules / GEMINI.md into a clean AGENTS.md.

These CLIs live under packages/ and are released by separate plans; this repo dogfoods agents-md-lint on its own templates in CI.

Contributing

Templates and matrix updates are the lifeblood of this repo. To add a stack, fix a command, or correct a tool's support status, read CONTRIBUTING.md and open an issue from one of our templates. Every PR is link-checked and content-linted automatically.

Credits & sources

• The AGENTS.md standard: https://agents.md/
• GitHub Blog — How to write a great agents.md: Lessons from over 2,500 repositories (Matt Nigh, 2025-11-25).
• ETH Zurich — Evaluating AGENTS.md (arXiv:2602.11988, Feb 2026), benchmark AGENTbench.
• Augment Code — AuggieBench golden-PR evaluation.
• philschmid — practical AGENTS.md length/structure guidance.

License

MIT © 2026 Taiizor