APP.md, keeps runtime behavior inside app/, and attaches local skills so agents can operate it safely.
What are Agent Applications?
Learn the package model, file roles, and why
APP.md stays central.Quick Start
Build and run your first Agent Application package in minutes.
Specification
Read the complete v1 format contract.
Package Authors
Write packages that stay portable, legible, and testable.
Runtime Implementers
Add Agent Applications support to your agent runtime or host tool.
Example App
Explore the reference to-do app that demonstrates the full contract.
Why Agent Applications?
Agent workflows become more reliable when the application — not the chat transcript — owns state and behavior. Agent Applications makes that contract explicit.- For package authors: document one stable CLI surface that agents and humans can inspect.
- For runtime implementers: load lightweight metadata first, then activate
APP.md, local skills, and app resources only when needed. - For teams: keep application instructions, command contracts, and safety rules in version-controlled files.
How it works
Create APP.md
Write a manifest that declares your CLI entrypoint, supported commands, scheduling behavior, and confirmation rules for destructive operations.
Implement your CLI
Put your runnable code inside
app/. The CLI must return machine-readable JSON by default and exit with a non-zero code on failure.Attach local skills
Add
SKILL.md files under skills/ to give agents context-aware operating guidance without changing the base app contract.