What Musketeer Is

Musketeer is a coordination and handoff discipline for working with multiple AI models on a single project. It is a way to structure thinking, doing, and checking so that each kind of cognitive work happens in the right place.

A discipline, not a tool

Musketeer is first and foremost a way of working. The CLI exists to support that way of working, but the discipline came first. The insight came from practice: different models are better at different kinds of thinking. When you use the wrong model for the wrong kind of work, you waste tokens, lose context, and erode trust in your own process.

The discipline is simple: separate the roles. Let one model hold the conversation and form intent. Let another model execute bounded tasks. Let a third model observe and report. Do not ask one model to do all three.

Coordination, not automation

Musketeer coordinates handoffs between models. It does not automate them. The human remains in the center, deciding when to hand off, what to hand off, and whether to accept what comes back.

This is deliberate. The goal is not to remove the human from the loop. The goal is to make the human's judgment more effective by ensuring that each model operates within its strengths.

Handoff, not orchestration

Orchestration implies a system that runs itself. Musketeer does not run itself. It prepares handoffs. It structures the information that moves between models. It records what was asked and what was delivered. But it stops before execution.

The handoff is the unit of work. A handoff includes intent, constraints, context, and acceptance criteria. When you hand off well, the receiving model has everything it needs. When you hand off poorly, you get back something you cannot use.

Human authority preserved

Every decision in Musketeer flows through the human. The Originator does not dispatch work automatically. The Executor does not verify its own output. The Cross-Examiner does not have the authority to block.

These are not limitations. They are design choices. The system is built for practitioners who want to remain in control of their work.

What Musketeer is not

It is not an agent framework. Agents imply autonomy. Musketeer does not grant autonomy to any model.

It is not a workflow engine. Workflow engines manage execution graphs. Musketeer manages cognitive separation.

It is not a runner. Runners execute tasks. Musketeer prepares tasks for execution elsewhere.

It is not autonomous. Nothing happens without the human choosing to make it happen.

It is not deterministic in the computational sense. The same inputs will not produce the same outputs. That is the nature of working with language models. Musketeer does not pretend otherwise.

The simplest description

Musketeer is a way to work with AI that respects the differences between models, preserves human judgment, and makes handoffs explicit. It is what you were already doing, named and structured.