Agent Context

Every project created by create-faster includes context for AI coding agents (Claude Code, Cursor, Codex, Gemini, Copilot). Instead of rediscovering the stack, conventions, and how libraries are wired, an agent reads it from the generated docs.

This is on by default. Opt out with --no-agent-context.

What gets generated

• AGENTS.md — the source of truth. Project header (stack, scripts, environment, conventions) plus a short section per selected library and addon.
• CLAUDE.md — a one-line @AGENTS.md import. Claude Code reads CLAUDE.md; every other tool reads the cross-tool AGENTS.md standard. No filesystem symlink, so it works on Windows, WSL, and Linux.

Single repo

my-project/
├── AGENTS.md     # project header + per-addon sections
└── CLAUDE.md     # @AGENTS.md

Monorepo (turborepo)

Each app gets its own context next to its code, while the root holds the project view (apps, packages, global environment).

my-project/
├── AGENTS.md            # project view: apps, packages, env scopes
├── CLAUDE.md            # @AGENTS.md
├── apps/web/AGENTS.md   # web's stack + libraries
├── apps/web/CLAUDE.md   # @AGENTS.md
└── apps/api/AGENTS.md   # api's stack + libraries

An app whose stack and libraries have no agent notes gets no AGENTS.md — there is never an empty file.

What the sections contain

Sections describe what is specific to your generated project — config location, scripts, environment variables, and conventions — and link out to official docs rather than re-explaining the libraries.

The project header is rendered from your composition (stack, ORM, database, linter, package manager, environment variables). Per-addon sections currently cover the stacks (Next.js, Expo, Hono) and the highest-value libraries and addons (Better Auth, shadcn/ui, tRPC, Drizzle, Prisma, PostgreSQL, MySQL). Addons without notes simply contribute no section.

Blueprints

Projects generated from a blueprint get extra context at the root AGENTS.md:

• ## About this project — states the project was scaffolded from a create-faster blueprint, links to the blueprint's documentation, and flags that the dummy data, fixtures, and example flows are placeholders meant to be replaced. The structure and integrations are the reusable part; the sample content is not.
• ## Architecture — how the blueprint is wired together.
• docs/agents/*.md — focused guides for individual aspects (for example, auth and permissions, or the data layer), linked from the Architecture section.

Opting out

bunx create-faster my-project --app my-project:nextjs --no-agent-context

When disabled, no AGENTS.md, CLAUDE.md, or docs/agents/ files are written.