What lands in a generated repo for agents

Every serious Arche preset aims to leave you with a repo that explains how to work on itself. This is the file-level checklist.

Canonical files

AGENTS.md

The entry map. Lists:

• Stack and package manager for this tree
• Commands that actually exist in package.json
• Pointers to .docs/ topics and deployment notes
• What not to load (archived plans, unrelated apps)

Other agent instruction files should symlink or point here—not fork the content.

.docs/

Internal topics—architecture, deployment, feature notes. Structured for agents and maintainers, not for marketing. When you change how auth or deploy works, update the nearest topic and the root map.

.plans/active/

One approved in-flight plan. Arche does not generate a pile of stale roadmaps. If you use plans, archive completed work to .plans/archive/ and keep a single active file.

arche.json

Machine-readable record of the scaffold command and options. Useful when you ask “how was this repo produced?” six months later.

What is intentionally missing

• .cursor/rules/ / .claude/rules/ duplication
• Claims like “production ready” without verification evidence
• Commands from the source template mixed into generated scripts

Keeping context accurate

Or run:

bun run dev:cli -- add agent-docs ./my-app

Verify against the preset

Support labels (Requires validation, etc.) describe the generator, not your fork. After heavy customization, your repo is yours—treat AGENTS.md as living documentation.

Deep dive: Agent context guide and CLI generated output.