EvaluationExperimental

my-virtual-world

Name: my-virtual-world
Rating: 3.0 (81 reviews)
Author: eliautobot

by eliautobot

Self-hosted 3D virtual world for reproducible multi-agent evaluation

JavaScript

Updated Aug 14, 2026

Stars

Forks

View on GitHub

Overview

Provides a self-hosted 3D virtual world to run agent harnesses like OpenClaw and Hermes for interactive testing. Hosts persistent scenes and object interactions so multiple agents can be exercised and observed in realistic, replayable scenarios. Includes hooks for harness telemetry and scripted event injection to reproduce multi-agent behaviors and failure modes. For structured harness usage, consider the Tool Use Pattern and deeper integration via the Agent-to-Agent Protocol (A2A).

Avoid common pitfalls

Learn what failures to watch for

Key Benefits

As agents become more autonomous and start interacting with each other, reproducible environments are essential to surface agent-to-agent failure modes and measure reliability. A shared virtual world makes agent track records comparable by standardizing scenarios, interactions, and logging. This repo fills a gap between synthetic benchmarks and real-world deployments by offering controllable, observable spaces for A2A evaluation and continuous agent evaluation. For related patterns, see the Event-Driven Agent Pattern.

Ideal For

Researchers and engineers who need a replayable, interactive environment to test agent-to-agent behaviors, failure modes, and evaluation harnesses. This aligns with the Human-in-the-Loop Pattern to ensure realistic oversight during experiments.

Real-World Examples

Reproduce and debug multi-agent system failures with replayable scenarios
Benchmark agent interactions and collect standardized telemetry for agent track records
Inject scripted events to stress-test delegation and observe cascading failures