AGENTS.md
AGENTS.md is a special file at the root of a branch workspace that provides the AI agent with context about your codebase. Think of it as a README for the agent — it tells the agent about your project’s purpose, architecture, conventions, and any other information that helps it work more effectively.
How AGENTS.md Works
Section titled “How AGENTS.md Works”When a conversation starts in a branch, Tarsk reads the AGENTS.md file from that branch’s directory and includes its content in the agent’s system prompt. Every message in that conversation has access to the project context you defined.
Edit AGENTS.md from Settings → Instructions (project scope, requires a selected branch), or place the file at the branch root manually.
Creating AGENTS.md
Section titled “Creating AGENTS.md”Place an AGENTS.md file at the root of your branch workspace:
## Overview
This is a Next.js e-commerce application with Stripe payments. The frontend usesReact Server Components with the App Router, and the API routes handle paymentprocessing and webhooks.
## Architecture
- `app/` — Next.js App Router pages and layouts- `app/api/` — API route handlers- `components/` — Shared React components- `lib/` — Utility functions and shared logic- `lib/db.ts` — Database client (Prisma)- `prisma/` — Prisma schema and migrations
## Conventions
- Use Server Components by default; only add "use client" when state or effects are needed- All API routes must validate input with Zod schemas- Database queries go through the Prisma client in `lib/db.ts`- Never access the database directly from componentsAuto-Generation
Section titled “Auto-Generation”If an AGENTS.md file does not exist when the agent first looks for it, Tarsk creates one with default template content. Edit it via Settings → Instructions or directly in the Files panel.
What to Include
Section titled “What to Include”The best AGENTS.md files are concise and focused on information the agent cannot discover by reading code alone:
Recommended Sections
Section titled “Recommended Sections”- Overview — What the project does, its purpose, and target users
- Architecture — Key directories, how the codebase is organized, and major subsystems
- Conventions — Coding standards, naming patterns, and rules the agent should follow
- Testing — How to run tests, what framework is used, where tests live
- Deployment — How the project is built and deployed
What Not to Include
Section titled “What Not to Include”- Information that is obvious from the code
- Large code snippets the agent can read directly from source
- Content that changes frequently and would become stale
Relationship to Rules
Section titled “Relationship to Rules”AGENTS.md and Rules both provide instructions to the agent, but they serve different purposes:
| AGENTS.md | Rules | |
|---|---|---|
| Purpose | Project context and architecture | Specific behavioral guidelines |
| Location | Branch root (AGENTS.md) | .agents/rules/*.md |
| Loading | Always included in system prompt | Conditionally loaded based on relevance |
| Scope | One file per branch | Multiple rules, each with its own focus |
Use AGENTS.md for broad project context and Rules for specific, actionable guidelines.
- Keep
AGENTS.mdunder 200 lines — the agent has a limited context window - Update it when the project’s architecture changes significantly
- Commit it to version control so your team benefits from the same agent context
- Use it to document decisions that are hard to reverse (e.g., “This project uses Prisma, not Drizzle”)