submerge12/agent-orchestration-harness

Agent Orchestration Harness, or AOH, is a system for running AI coding agents in a way that stays visible and controllable rather than letting them work unsupervised. Instead of just firing off a prompt and hoping for a good result, it turns a piece of work into a structured task with clear acceptance criteria, breaks that task into smaller steps that depend on each other, and hands each step to an interchangeable agent such as PI Agent, Codex, or Claude Code through one shared interface. As the agents work, AOH enforces limits on what they are allowed to touch, keeps their workspaces separate from each other, tracks how much budget they are using, and pauses for a human decision at points that matter rather than every small action. When a step is claimed finished, the system checks the real changes and evidence the agent produced instead of just trusting its report, and an independent review decides whether to accept the work, ask for repairs, block it, or escalate it to a person. All of this activity is recorded in an append only history so the whole process can be replayed later, and a dashboard called the Control Room lets you inspect projects, tasks, blockers, evidence, and decisions. Getting started requires Node.js 22 or newer, after which you install dependencies and run the included type checks and test suite. The project states plainly which parts are working today, such as the governance rules, the agent adapters, and the review flow, and which features, like natural language task dispatch or long running resumable sessions, are not yet shipped. The README does not state a specific license, only that license terms are defined somewhere in the repository.