Introduction

xcaffold is a Harness-as-Code toolchain that lets you declare AI agents, skills, rules, and behaviors in a single .xcaf YAML file and compile them into any provider's native format. Write your agent configuration once and target Claude Code, Cursor, Antigravity, GitHub Copilot, or Gemini CLI — xcaffold handles the translation deterministically.

There are two distinct products. Choose your track:

---

xcaffold CLI — Open Source (Apache 2.0)

The xcaffold CLI is a standalone Go binary. It runs entirely on your local machine, requires no account, and is free forever.

What it does:

• Compiles project.xcaf into each provider's native output format
• Detects manual drift with a SHA-256 lockfile (project.xcaf.state)
• Validates compiled output against project policies
• Simulates and evaluates agent behavior (xcaffold test --judge)

Supported providers and their output directories:

Who it's for: Individual contributors, open-source maintainers, any developer who wants version-controlled, reproducible AI agents across one or more providers.

→ Start here: Quick Start

---

xcaffold Platform — Cloud (Team & Enterprise)

The xcaffold Platform is a SaaS layer that adds team collaboration, state synchronization, and fleet telemetry on top of the CLI. The CLI is still the source of truth; the Platform adds visibility and control across your organization.

What it adds:

• Centralized project.xcaf.state sync — no drift conflicts between developers
• Fleet Telemetry — Agents Under Management (AUM) dashboards
• GitOps Dispatch Bay — trigger xcaffold apply from GitHub PRs
• Policy-as-Code enforcement (Enterprise tier)

Who it's for: Engineering managers, platform teams, and organizations running AI agents across multiple repositories and providers.

→ Start here: Platform Overview

---

The Lifecycle

Both products share the same compilation lifecycle. The CLI executes each phase locally; the Platform provides remote visibility and enforcement around the same pipeline.