> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tryvoss.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Orchestration overview

> How Voss runs AI agents as a constrained engineering organization — six primitives, a board, a session tree, declared roles, and independent review.

The harness gives one agent a bounded loop. Orchestration extends the same boundaries to a whole team: declared roles, partitioned budget, scoped tools, independent review, and a replayable trail.

The promise:

> Voss lets AI agents work like an engineering team while keeping every action scoped, budgeted, reviewed, and replayable.

## The six primitives

Every run is built from the same six primitives.

| Primitive     | Meaning                                             | Where it lives                                              |
| ------------- | --------------------------------------------------- | ----------------------------------------------------------- |
| Capabilities  | The agent toolbelt — typed, permissioned, auditable | [`voss capabilities`](/reference/capabilities)              |
| Principles    | Engineering culture injected into every agent       | [`.voss/principles.yml`](/orchestration/principles)         |
| Orchestration | Delegation and integration                          | [Engineering Manager loop](/orchestration/em-loop)          |
| Roles         | Specialist lenses with their own scope and budget   | [`.voss team{}`](/orchestration/team-config)                |
| Memory        | Institutional knowledge across runs                 | [Session tree](/orchestration/session-tree), project memory |
| Verification  | The independent review loop                         | [Reviewer A/B](/orchestration/reviewers)                    |

## One run, end to end

```text theme={"theme":"github-dark"}
human idea
  -> EM derives cards
  -> assign roles from the declared roster
  -> partition budget down the session tree
  -> dispatch workers in parallel (within WIP limits)
  -> Reviewer-A authors the bar from the idea
  -> Reviewer-B judges narrative-blind
  -> board gates Done on independent review
  -> RunFinal record + review sidecars persisted
```

## The MVP flow

A team is declared once in `.voss`, then a single goal runs as scoped, reviewed work.

```bash theme={"theme":"github-dark"}
voss team check
voss team run "Add password reset flow with tests"
voss board
voss review latest
```

This run proves the core invariants:

* The EM cannot invent roles outside the declared roster.
* Workers cannot write outside their assigned scope.
* Budget cannot be oversold — it is enforced down the session tree.
* Agents cannot mark their own work Done.
* Done requires independent reviewer evidence.

<CardGroup cols={2}>
  <Card title="Team config" icon="users" href="/orchestration/team-config">
    Declare roles, scope, budget, and tools in `.voss team{}`.
  </Card>

  <Card title="Engineering Manager loop" icon="workflow" href="/orchestration/em-loop">
    The constrained tech lead that decomposes and integrates.
  </Card>

  <Card title="Board" icon="kanban" href="/orchestration/board">
    Columns, WIP limits, and artifact-gated transitions.
  </Card>

  <Card title="Reviewers" icon="scale" href="/orchestration/reviewers">
    The independent Reviewer-A / Reviewer-B split.
  </Card>

  <Card title="Session tree" icon="git-branch" href="/orchestration/session-tree">
    Per-agent budget fan-out and durable nodes.
  </Card>

  <Card title="Principles" icon="shield-check" href="/orchestration/principles">
    Engineering culture as first-class config.
  </Card>
</CardGroup>