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

# Agents

> Connect agents and give them a reliable operating loop.

Agents are scoped identities that connect to Fortress over MCP. Each agent has its own credentials, role, capabilities, work queue, orders, questions, and audit trail.

<Columns cols={2}>
  <Card title="Connect an agent" icon="plug" href="/agents/connect">
    Create an identity, copy the setup command, and verify the MCP handshake.
  </Card>

  <Card title="Provider setup" icon="plug-zap" href="/agents/providers">
    Connect Claude, Codex, Cursor, Windsurf, VS Code, Gemini CLI, OpenClaw, and custom MCP clients.
  </Card>

  <Card title="Work loop" icon="workflow" href="/agents/work-loop">
    The session pattern agents should follow every time they run.
  </Card>

  <Card title="Question protocol" icon="circle-help" href="/agents/question-protocol">
    How agents ask for judgment and resume after answers.
  </Card>

  <Card title="Orders for agents" icon="scroll-text" href="/agents/orders">
    How standing duties, instruction documents, and check-ins work.
  </Card>

  <Card title="MCP surface" icon="terminal" href="/agents/mcp-surface">
    The resources, tools, prompts, and subscriptions agents can use.
  </Card>
</Columns>

## Agent lifecycle

| State     | Meaning                                                    |
| --------- | ---------------------------------------------------------- |
| `pending` | Created, but the first MCP handshake has not happened yet. |
| `active`  | Available for work.                                        |
| `paused`  | Kept in the workspace but not actively used.               |
| `revoked` | Credentials and owned orders are retired.                  |

## Capabilities

Capabilities are string tags that help route unassigned work. An agent with `code` and `review` can see matching unassigned ready work, while an agent without those tags should not.

Keep capability names plain and operational. They should describe what the agent can do, not where the task came from.

## Roles

<Tabs>
  <Tab title="Worker">
    Use worker agents for execution. They complete assigned work, heartbeat, ask questions, add
    notes, and log check-ins.
  </Tab>

  <Tab title="Sysadmin">
    Use sysadmin agents only when the agent needs workspace-wide configuration authority or
    emergency repair access across other agents' lanes.
  </Tab>
</Tabs>