Quick Start

This guide walks you through using Assemble for the first time. In 5 minutes, you’ll understand how agents collaborate to deliver real work.

Step 1 — Set up a project

mkdir my-app && cd my-app
npm init -y
npx cohesiumai-assemble

Select Cursor and Claude Code as platforms, choose the startup profile, and enable YOLO mode for autonomous execution.

Step 2 — Open your AI coding tool

Open the project in Cursor, or start Claude Code in the terminal:

claude

Your AI tool automatically reads the generated configuration files. You now have access to all 34 agents.

Step 3 — Try your first command

/go create a REST API for user management with authentication

Jarvis assesses this as a COMPLEX task and activates the Spec-Driven methodology:

@professor-x produces spec.md with requirements
@tony-stark produces plan.md with architecture decisions
@captain-america produces tasks.md with the breakdown
@bruce-banner implements the API endpoints
@doctor-strange designs the database schema
@hawkeye writes the tests

Each agent’s deliverables appear in assemble-output/.

Step 4 — Use shortcuts

For common tasks, use workflow shortcuts:

/feature add password reset to the auth flow
/bugfix the login endpoint returns 500 on invalid email
/review check the PR for the payment module

Each shortcut triggers a predefined workflow with the right agents in the right order.

Step 5 — Enable YOLO mode

With YOLO mode, agents execute autonomously without asking for validation at each step:

yolo: true

Three levels are available:

YOLO — agents work autonomously, stop only for production actions or missing info
Hardcore (/yolo-hardcore) — full autonomy, stops only for production
Full (/yolo-full) — complete autonomy, stops for nothing

Step 6 — Party mode

Want multiple agents to brainstorm together?

/party Should we use microservices or a monolith for this project?

Jarvis summons relevant agents (plus Deadpool as the contrarian) for a collaborative debate. Each agent speaks in character and challenges the others.

Example commands

Command	What happens
`/go fix the search performance issue`	Jarvis routes to relevant agents
`/feature add dark mode toggle`	Feature development workflow
`/bugfix auth token expires too early`	Bug fix workflow
`/review`	Code review pipeline
`/sprint plan the next 2-week sprint`	Sprint cycle workflow
`/release prepare v2.1.0`	Release cycle workflow
`/party how should we handle rate limiting?`	Multi-agent brainstorm
`@spider-man refactor the dashboard components`	Direct agent access

What’s next?

Agents — understand the 34 specialized agents
Workflows — learn about the 15 predefined workflows
YOLO Mode — configure autonomous execution
Custom Agents — create your own agents