Skip to content
AOF — Agentic Ops Fabric

Quick Start Tutorial

This tutorial walks you through AOF’s core workflow: defining an org chart, dispatching a task, and completing it through a workflow gate. It takes about 10 minutes.

Before You Start

Make sure you’ve completed Installation & Setup and have AOF initialized in a project directory.

Step 1: Create Your Org Chart

The org chart is the foundational declaration of your agent swarm. Create org/org-chart.yaml in your project:

org/org-chart.yaml
schemaVersion: 1


agents:
  - id: swe-backend
    name: Backend Engineer Agent
    active: true
    capabilities:
      - typescript
      - nodejs
      - apis
    team: engineering


  - id: swe-reviewer
    name: Senior Engineer (Reviewer)
    active: true
    capabilities:
      - typescript
      - architecture
      - code-review
    team: engineering


teams:
  - id: engineering
    name: Engineering
    members:
      - swe-backend
      - swe-reviewer


roles:
  developer:
    agents:
      - swe-backend
    description: "Implements features and fixes"


  reviewer:
    agents:
      - swe-reviewer
    description: "Reviews code quality and correctness"


memoryPools:
  hot:
    path: memory/hot
  warm:
    - id: per-agent
      path: memory/warm/agents
  cold:
    path: memory/cold

Validate it:

Terminal window
aof org validate
# ✅ Org chart valid: 2 agents, 1 team, 2 roles

Create a workflow that enforces your development process. For this tutorial, we’ll use a simple implement → review cycle:

org/workflows/simple-review.yaml
workflow:
  name: simple-review
  rejectionStrategy: origin


  gates:
    - id: implement
      role: developer
      description: "Implement the feature"


    - id: review
      role: reviewer
      canReject: true
      description: "Review and approve the implementation"


  outcomes:
    complete: advance
    needs_review: reject
    blocked: hold

Step 3: Dispatch Your First Task

Dispatch a task using the CLI (or the aof_dispatch agent tool — same semantics):

Terminal window
aof task create "Add rate limiting to the auth endpoint" \
  --agent swe-backend \
  --priority high

Output:

✅ Created task: TASK-2026-02-21-001
   Title: Add rate limiting to the auth endpoint
   Priority: high
   Status: backlog
   Agent: swe-backend
   Path: tasks/backlog/TASK-2026-02-21-001.md

The task file looks like:

---
schemaVersion: 1
id: TASK-2026-02-21-001
title: Add rate limiting to the auth endpoint
status: backlog
priority: high
routing:
  agent: swe-backend
createdAt: 2026-02-21T15:00:00Z
updatedAt: 2026-02-21T15:00:00Z
createdBy: cli
dependsOn: []
---

Step 4: Promote the Task to Ready

Move it from backlog to ready so the scheduler can dispatch it:

Terminal window
aof task promote TASK-2026-02-21-001

Or set its status directly:

Terminal window
aof task status TASK-2026-02-21-001 ready

Step 5: Run the Scheduler

The scheduler picks up ready tasks and dispatches them to their assigned agents:

Terminal window
# Dry-run (see what would happen, no state changes)
aof scheduler run


# Active mode (actually dispatches)
aof scheduler run --active

The scheduler output shows:

Scheduler poll — 1 ready task found
  → Dispatching TASK-2026-02-21-001 to swe-backend
  → Status: in-progress, Lease: 30m
✅ Dispatched 1 task

The task file is now at tasks/in-progress/TASK-2026-02-21-001.md with a lease:

---
status: in-progress
lease:
  agent: swe-backend
  expiresAt: 2026-02-21T15:30:00Z
  sessionId: session-abc123
---

Step 6: Agent Completes the Task

When the swe-backend agent finishes its work, it calls aof_task_complete (via the MCP tool or OpenClaw gateway):

{
  "tool": "aof_task_complete",
  "params": {
    "taskId": "TASK-2026-02-21-001",
    "summary": "Added token bucket rate limiting (100 req/min) to POST /auth/login. Included tests for limit enforcement and header passthrough.",
    "actor": "swe-backend"
  }
}

For tasks without a workflow gate, this moves the task to done.

Step 7: Using Workflow Gates

Let’s dispatch a gated task. First, create the task with a workflow reference:

Terminal window
aof task create "Implement user session expiry" \
  --agent swe-backend \
  --priority normal

Then edit the task file to add the workflow:

routing:
  agent: swe-backend
  workflow: simple-review

When the scheduler dispatches this task, it routes to the implement gate. The swe-backend agent works on it and calls:

{
  "tool": "aof_task_complete",
  "params": {
    "taskId": "TASK-2026-02-21-002",
    "outcome": "complete",
    "summary": "Session expiry implemented. 15-minute timeout with sliding window. Tests added.",
    "actor": "swe-backend",
    "callerRole": "developer"
  }
}

The gate evaluator sees outcome: "complete" and advances the task to the review gate. It’s now dispatched to swe-reviewer.

The reviewer approves:

{
  "tool": "aof_task_complete",
  "params": {
    "taskId": "TASK-2026-02-21-002",
    "outcome": "complete",
    "summary": "LGTM. Good test coverage, clean implementation.",
    "actor": "swe-reviewer",
    "callerRole": "reviewer"
  }
}

Task moves to done. ✅

Or the reviewer rejects with feedback:

{
  "tool": "aof_task_complete",
  "params": {
    "taskId": "TASK-2026-02-21-002",
    "outcome": "needs_review",
    "blockers": ["Missing test for concurrent session invalidation", "Token not invalidated on password change"],
    "summary": "Needs fixes before approval.",
    "actor": "swe-reviewer",
    "callerRole": "reviewer"
  }
}

The task loops back to implement for swe-backend to address the feedback.

Step 8: View Your Board

Terminal window
aof board
┌─────────────────────────────────────────────────────┐
│  AOF Kanban — Engineering                           │
├──────────┬──────────┬─────────────┬────────────────┤
│ Backlog  │  Ready   │ In-Progress │    Done        │
├──────────┼──────────┼─────────────┼────────────────┤
│          │          │ TASK-001    │ TASK-002       │
│          │          │ Rate limit  │ Session expiry │
│          │          │ swe-backend │ [reviewed]     │
└──────────┴──────────┴─────────────┴────────────────┘

Next Steps