Agents

The agent system

OMX ships 33 role-specific agents. Each agent keeps a narrow responsibility, and the orchestrator chooses among them based on the shape of the task.

In other words, OMX does not rely on one general-purpose worker to do everything. It separates exploration / planning / implementation / verification / review into distinct roles.

Four-lane structure

The agents are grouped into four lanes.

Build / Analysis

This is the center of the development flow. It runs from discovery to planning, implementation, and verification.

This lane is the default execution line from “what are we building?” to “is it actually done?”.

Review

This lane checks quality and risk before changes should be considered complete.

This lane is less about “does it run?” and more about is it good enough to trust and ship?

Domain

These are specialists called in when the work needs narrower expertise.

For example:

• test-engineer — test strategy
• designer — UI and UX work
• writer — docs and user guidance
• researcher — external documentation and reference research
• dependency-expert — SDK / package decisions
• git-master — Git history and commit hygiene

This lane does not replace the default execution flow. It strengthens it where specialized expertise is needed.

Coordination

These agents challenge, coordinate, and tighten what the other lanes produce.

For example:

• critic — challenge plans and designs
• team-executor — support team execution lanes
• quality-strategist — evaluate quality strategy

This lane is less about direct implementation and more about making the overall process safer and sharper.

How do you choose an agent?

A simple mental model is:

So in practice, most workflows begin in Build / Analysis, and then pull in review, domain, or coordination roles when needed.