Openclaw Multiagent

This is a submission for the OpenClaw Challenge.

What I Built

What if you want to build something but you don’t even have access to your laptop?

I built a workflow where you can create a Pull Request directly from Telegram, trigger tests, and request code review — all without opening GitHub or your IDE.

The system uses Telegram as the interface and agents behind the scenes to execute development tasks. Instead of manually writing code, pushing commits, and opening PRs, you can just send a command in Telegram and the agents handle the rest.

Key capabilities:

Discuss Architecture before making any changes
Create PRs from Telegram commands
Automatically run repository tests
Request reviewers
Track PR status

This effectively turns Telegram into a remote development assistant.

How I Used OpenClaw

OpenClaw powers the automation layer behind the system.

I designed a multi-agent workflow where each agent is responsible for a specific development task. These agents invoke OpenCode sessions to interact with the repository, analyze the codebase, and execute development actions.

Instead of a single AI doing everything, the system uses specialized agents coordinated through an orchestrator, making the workflow more modular and scalable.

Agents in the System

1. Code Agent

Generates and edits code
Implements features and fixes bugs
Writes and updates documentation
Applies architectural patterns

1. Plan Agent

Breaks down complex tasks into smaller steps
Creates execution strategies
Validates feasibility of approaches
Routes tasks to specialized agents

1. Review Agent

Analyzes code for quality issues
Identifies anti-patterns and potential bugs
Provides improvement suggestions
Validates code against best practices

1. Debug Agent

Investigates errors and failing workflows
Traces issues within the codebase
Suggests fixes and workarounds
Validates fixes by running tests

1. Architect Agent

Designs system architecture
Defines component interfaces
Plans migrations and refactors
Ensures scalability and maintainability

1. Orchestrator Agent (hs-maestro)

Coordinates multi-step workflows
Manages collaboration between agents
Tracks progress across tasks
Handles retries and recovery when failures occur
Workflow

When a command is triggered from Telegram, the Orchestrator Agent coordinates the workflow:

The Plan Agent breaks down the request.
The Code Agent generates or modifies code.
The Review Agent analyzes the changes.
The Debug Agent resolves failures if tests break.
The Architect Agent ensures the solution aligns with system design.

Each step is executed through OpenCode sessions powered by OpenClaw, allowing agents to safely interact with the repository and development environment.

Demo

ayushanandbtcs / openclaw-agents

OpenClaw Multi-Agent System

Real agent definitions powering the automation workflow.

Multi-Agent Architecture

                    ┌─────────────────────────────────┐
                    │         USER REQUEST            │
                    └─────────────┬───────────────────┘
                                  │
                                  ▼
                    ┌─────────────────────────────────┐
                    │      HS-MAESTRO                 │
                    │   Orchestrates entire workflow  │
                    │   Dispatches to agents          │
                    └─────────────┬───────────────────┘
                                  │
                    ┌─────────────┴─────────────┐
                    │                           │
                    ▼                           ▼
        ┌─────────────────────┐   ┌─────────────────────┐
        │   PLANNER AGENT     │   │   MYCROFT AGENT     │
        │   Creates plan      │   │   Researches/       │
        │   Decomposes tasks  │   │   Investigates      │
        └──────────┬──────────┘   └─────────────────────┘
                   │
                   ▼
        ┌─────────────────────┐
        │   CODER AGENT       │
        │   Implements code   │
        │   Writes tests      │
        └──────────┬──────────┘
                   │
                   ▼
        ┌─────────────────────┐
        │   REVIEWER AGENT    │
        │   Checks quality    │
        │   Finds issues      │
        └──────────┬──────────┘
                   │
                   ▼
        ┌─────────────────────┐
        │   CRITIC AGENT      │◄── "Prove it to me"
        │   Gatekeeper        │    "Show me evidence"
        │   Blocks until      │    "Why not alternative?"
        │   convinced         │
        └──────────┬──────────┘
                   │ (approved)
                   │
                   ▼
        ┌─────────────────────┐
        │   BACK TO MAESTRO   │
        │   Create PR         │
        └─────────────────────┘

The Agents

1. HS-Maestro Agent (Orchestrator) ⭐ FIRST

…

View on GitHub

What I Learned

ClawCon Michigan