Builder Handbook

You are one of the builders shortlisted for Forge Sprint 02, the multi-agent AI development sprint. On Saturday 20 June 2026 you will spend six hours building a working system where multiple AI agents coordinate through Slack to solve a real business problem, using the exact methodology that has taken production teams from 15 PRs per month to 1,000.

What is at stake

• Champion (1st): Cash prize (revealing soon) plus NMG offer plus Certificate of Participation plus NMG Labs Skool membership plus guaranteed Forge 3 seat with dedicated mentor.
• Runner-up (2nd): Cash prize (revealing soon) plus NMG offer plus Certificate plus NMG Labs Skool plus guaranteed Forge 3 seat.
• 3rd Place: Cash prize (revealing soon) plus NMG offer plus Certificate plus NMG Labs Skool plus guaranteed Forge 3 seat.
• Top 10: Certificate plus NMG Labs Skool membership plus Forge 3 priority registration.
• All Participants: Certificate of Participation plus NMG Labs Skool membership plus standing invitation to future sprints.

The Forge 2 to Forge 3 pipeline

Forge 2 is the qualifying round for Forge 3 (ProcessHQ Hackathon, Jul 4). The multi-agent skills you build here are exactly what you need to solve real product development challenges for ProcessHQ.AI, a live SaaS platform with 56+ enterprise demos and paying customers. Forge 3 prizes are significantly larger.

What we test

We test your ability to:

• Configure and orchestrate multiple AI agents with distinct roles.
• Use Slack as a communication and coordination layer (human-in-the-loop).
• Route tasks to the right model (expensive for planning, cheap for execution).
• Build automated quality gates (CI/CD, testing, health checks).
• Ship a working system that produces real, useful output.

There is no cheat sheet on Sprint Day. You must arrive having studied the entire curriculum in Section 07. The challenge brief is revealed at the venue.

We picked this stack because it mirrors a proven production setup: Hermes for orchestration, OpenClaw for coding, Slack for communication.

Model routing (encouraged)

The Rule: use the most expensive model for architecture decisions, cheap models for execution.

Permitted alternatives

• Claude Pro / Anthropic API credits: allowed. Declare on registration. Free-stack builders get tie-break advantage.
• 32 GB RAM laptop plus Ollama: run everything locally with larger models. Fully supported.
• Additional agents beyond Hermes plus OpenClaw: encouraged. Add testing agents, documentation agents, deployment agents.

Do this before Sprint Day. Arrive with a verified, working Hermes Agent.

Step 1 · Install prerequisites

Check the official Hermes Agent quickstart for the latest installation commands. As of June 2026:

# Node.js 22+ required
node --version    # must be v22.0.0 or higher
git --version     # any recent version

# Install Hermes Agent (verify exact command at hermes-agent.nousresearch.com/docs)
npx hermes-agent@latest init my-forge-agent
cd my-forge-agent

Step 2 · Configure your model

Edit the Hermes Agent config to use your Groq API key:

# Set Groq as your provider
export GROQ_API_KEY="your-groq-api-key-here"

Configure hermes.config.json:

{
  "model": {
    "provider": "groq",
    "model": "moonshotai/kimi-k2-instruct",
    "apiKey": "${GROQ_API_KEY}"
  },
  "memory": {
    "enabled": true,
    "persistence": "local"
  },
  "skills": {
    "autoLearn": true,
    "directory": "./skills"
  }
}

Step 3 · Verify Hermes works

hermes chat "What is 2+2? Then create a file called test.txt with the answer."
# Should respond AND create the file

Step 4 · Create your first skill

Create skills/hello-world/SKILL.md:

# Hello World Skill

## Trigger
When asked to greet someone or say hello.

## Steps
1. Read the person's name from the input
2. Generate a personalized greeting
3. Save the greeting to a file called greetings.md

Test it:

hermes chat "Say hello to the NMG Labs team"
# Should use the skill automatically

Resources

• Hermes Agent Quickstart
• Hermes Agent GitHub
• Hermes Agent Documentation

Step 1 · Install OpenClaw

Check the official OpenClaw docs for the latest installation commands. As of June 2026:

# Install OpenClaw (verify exact command at docs.openclaw.ai)
npm install -g openclaw

# Or via npx
npx openclaw@latest onboard

The onboarding wizard guides you through setting up the gateway, workspace, channels and skills.

Step 2 · Configure model provider

# Set up with Groq (same key as Hermes)
openclaw config set provider groq
openclaw config set apiKey YOUR_GROQ_API_KEY
openclaw config set model moonshotai/kimi-k2-instruct

Step 3 · Configure Slack channel

# Connect OpenClaw to your Slack workspace
openclaw channel add slack
# Follow the OAuth prompts to connect

Step 4 · Verify OpenClaw works

openclaw chat "Create a Python script that prints 'Hello from OpenClaw'"
# Should generate and optionally run the script

Step 5 · Create agent identity

Create IDENTITY.md in your OpenClaw workspace:

# Coding Agent Identity

You are a coding agent specialized in writing clean, tested code.
You receive tasks from the Orchestrator (Hermes Agent) via Slack.
For every task:
1. Acknowledge receipt in Slack
2. Write the code
3. Run tests
4. Report status: What I Did / What Failed / What Needs Review

Resources

• OpenClaw Documentation
• OpenClaw GitHub
• OpenClaw Onboarding Guide

This is the critical piece. Rule #4: Bots never communicate directly. All communication goes through Slack.

Step 1 · Create a free Slack workspace

Go to slack.com/get-started and create a new workspace. Name it something like forge-02-[yourname].

Step 2 · Create these channels

Step 3 · Connect Hermes to Slack

Follow the Hermes Slack Integration Guide to create a Slack app and connect Hermes.

Step 4 · Connect OpenClaw to Slack

OpenClaw has built-in Slack channel support. Connect it to the same workspace.

Step 5 · Test agent communication

1. Ask Hermes to post a message in #sprint-main: "Sprint started. Assigning tasks."
2. Ask OpenClaw to post in #agent-coder: "Ready to receive tasks."
3. Verify you can see both messages in Slack.
4. Try having Hermes assign a task that OpenClaw picks up.

The communication pattern you must demonstrate

Hermes (in #sprint-main): "Task 1 assigned to @coder: Write a function to parse CSV files."
OpenClaw (in #agent-coder): "Task 1 received. Working on it."
OpenClaw (in #agent-coder): "Task 1 complete. PR #1 submitted. Tests passing."
Hermes (in #sprint-main): "Task 1 verified. Moving to Task 2."
You (in #human-review): "@hermes approved. Merge PR #1."

Hardware

• ☐Your own laptop with at least 8 GB RAM (16 GB recommended for local models).
• ☐Charger and power adapter.
• ☐Government photo ID (Aadhaar, College ID, Passport or Driving Licence).
• ☐Headphones (recommended).

Software installed and verified

• ☐Node.js v22.0.0+, verified with node --version.
• ☐Git, verified with git --version.
• ☐Hermes Agent, verified with hermes --version.
• ☐OpenClaw, verified with openclaw --version.
• ☐VS Code or your preferred editor.

Accounts created

• ☐GitHub account with at least one public repo.
• ☐Groq Console account at console.groq.com with API key.
• ☐Slack workspace created with 5 channels (see Section 05).
• ☐Hermes connected to Slack (tested).
• ☐OpenClaw connected to Slack (tested).
• ☐Loom account (free) at loom.com.
• ☐(Optional) Ollama installed for local model fallback.

Pre-Sprint test

Before Sprint Day, build this mini-project to verify your entire stack works:

If this works end-to-end with visible Slack messages, you are Sprint-ready.

There is no cheat sheet. You must self-study all 7 modules before Sprint Day.

Challenge structure (revealed at 09:50 AM)

On Sprint morning, you receive:

1. A real business problem with a dataset.
2. A set of constraints (for example "Must use at least 3 agents," "Must route models strategically").
3. A surprise second dataset released at 17:00 for the auto-tester.

Architecture requirements (mandatory for scoring)

Your system MUST include:

• ☐Hermes Agent as the orchestrator (plans, decomposes, assigns, tracks).
• ☐OpenClaw as at least one coding agent (writes code, runs tests).
• ☐Slack channels showing all agent communication (minimum 3 channels).
• ☐Automated quality gate (tests, health check or validation script).
• ☐Human approval step before final output (you review and approve in Slack).

Required outputs

1. Working system that produces correct results on the challenge dataset.
2. Public GitHub repo with incremental commits and agent-log.md.
3. Slack export or screenshots showing agent communication.
4. README documenting architecture (which agent does what, which model, why).
5. Loom video (max 5 minutes) walking through the system.
6. Architecture diagram (can be hand-drawn, Mermaid or any format).

Free-stack advantage

Eligibility

Open to 3rd/4th year UG, PG and recent graduates (within 12 months) across India. Forge 1 attendance is NOT required. Fresh builders welcome. Individual competition only.

8 disqualification triggers

1. Private repo at deadline.
2. Missing agent-log.md or Slack evidence.
3. No Hermes Agent in architecture (must be the orchestrator).
4. Single giant commit instead of incremental build phases.
5. Auto-tester failure within 5 minutes.
6. Code reading score of 0 (cannot explain any block of your code).
7. 75%+ code duplication with existing public repositories.
8. Conduct violation (harassment, plagiarism, dishonesty).

Intellectual property

Your code, your IP. NMG retains a non-exclusive license to publish a case study (with credit to you). You are free to commercialize, open-source or fork after. Full detail lives on our Terms and Conditions and Code of Conduct pages.

Questions? Forge Captain Ayush Gupta · email labs@nmgdigital.com (response under 24 hours).
See the Forge 2 sprint page, Terms and Code of Conduct.