What is stax?

Spec Version 1.0.0

This documentation covers stax spec version 1.0.0. The specification is actively being developed. All compiled JSON payloads include specVersion: "1.0.0".

stax is a distribution standard for AI agents. It defines how to describe, bundle, version, validate, verify, and distribute agent artifacts as OCI artifacts.

stax artifacts are the agent equivalent of docker build, docker push, and docker pull: authors build and publish immutable agent artifacts, and consumers pull, inspect, verify, materialize, install, or import them into the environments where agents run.

stax does not run agents, orchestrate them, or manage their lifecycle. Those concerns belong to runtimes, orchestrators, IDEs, CLIs, hosted agent platforms, and cloud systems built on top of the format.

stax is to agents what the OCI Image Spec is to containers, what package.json is to JavaScript packages, and what Chart.yaml is to Helm charts.

Philosophy

Why agents need a distribution standard and the core thesis behind stax

Architecture

Core objects, the seven jobs, and conformance levels

Roadmap

Implementation status and what's coming next

Current Focus

In the current market, stax is best understood first as a standard for coding-agent asset distribution. The broader vision still applies, but the most mature current fit is for:

Instructions and instruction trees
Subagents
Skills
Rules
MCP configuration
Shared source context
Approvals, policy, and promotion metadata

The Core Thesis

Software won because it became packageable.

Containers became standard once there was a portable image format. Kubernetes became practical once workloads could be described declaratively. Package ecosystems scaled once artifacts were versioned, signed, addressable, and installable.

Agents are still early because most of the ecosystem is stuck in pre-package form:

Prompts in Markdown
Skills in folders
MCP config in runtime-specific JSON
Runtime-specific instruction files
Ad hoc memory directories
Undocumented assumptions about tools, sources, and environment

The stax insight

Today an "agent" is usually a pile of files tied to one tool. stax turns that into a real artifact — versioned, portable, verifiable, and distributable.

What stax Is

stax is the distribution layer for agents. It is the standard for:

Authoring

Define agent artifacts with a TypeScript-first SDK

Bundling

Package the canonical brain: persona, prompt, skills, rules, MCP, knowledge

Dependencies

Express portable dependencies on packages and source artifacts

Packaging

Build reusable agent packages for shared behavior and context

Publishing

Push to OCI registries with versioning and content addressing

Verification

Sign and verify supply chain integrity with Sigstore

Promotion

Promote artifacts across dev, staging, and production environments

Materialization

Translate artifacts into runtime-native formats for any target

Discovery

Search, install, update, and roll back agent artifacts

What stax Is Not

stax is intentionally not:

An agent runtime
An orchestrator
A scheduler
A workflow engine
A session manager
An execution loop
A hosting platform
A queue processor
A secret manager
An observability system
A multi-agent topology engine

Those products may consume stax, extend stax, or integrate with stax. They should not be collapsed into stax itself.

The Standard stax Should Become

OCI for agent artifacts
npm for reusable agent packages
Helm-like for installable agent bundles
Sigstore-aware for agent trust
A universal import/export layer between agent runtimes