Developer

Build agents and skills against the Forge runtime, or bring your own Claude Agent SDK service and govern it with the Agent SDK — the developer-facing documentation, in one place with the platform docs.

Every agent the platform builds runs on Forge, the open agent runtime: skills, tools, channels, egress policy, audit — the runtime is where all of it executes. This section carries the runtime's complete documentation so you can build skills, run agents locally, and debug production behavior without leaving the platform docs. It also covers the Agent SDK — the initializ governance layer and A2A authoring kit for agents you build on the Claude Agent SDK in your own repository and deploy from CI.

Start here depending on what you're doing:

Agent SDK

Govern a Claude Agent SDK service with zero code changes — the loader and CI governance wrapper — and expose typed skills over A2A with @initializ/a2a-kit.

Getting started

Install the runtime, talk to a working agent in under a minute, and take your first skill to production.

Core concepts

How the runtime works: skills, tools and builtins, channels, memory, scheduling, hooks, and observability.

Writing skills

The SKILL.md format, custom skills, the skills CLI, and contributing skills back.

Reference

The forge.yaml schema, CLI reference, environment variables, library modules, and integration surfaces.

Security

The runtime's enforcement layers: egress control, guardrails, deferred decisions, audit signing, and the trust model.

MCP

Connect agents to MCP servers: configuration, delegated consent, audit events, and troubleshooting.

These pages are synced verbatim from the runtime's own documentation — they are regenerated by tooling, so treat them as read-only here. How the runtime concepts surface for platform users (deployed agents, console screens) is distilled separately under Agent runtime.