Harness Arena — AI skill for Claude Code
Blind, LMSYS-style arena for comparing AI coding agent harnesses (Claude Code, Codex CLI, OpenClaw, Hermes, OnDemand) instead of models.
How to install Harness Arena
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open Ondemand-OSS/harness-arena and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Harness Arena does
Blind, LMSYS-style arena for comparing AI coding agent harnesses (Claude Code, Codex CLI, OpenClaw, Hermes, OnDemand) instead of models. Same task, same model, isolated runs , humans judge blind, results roll up into an Elo leaderboard. FastAPI + MongoDB backend.
Alternatives in AI
- Gensee Crate — Run AI harnesses like Claude Code, Codex, Copilot, Antigravity, and Omnigent in disposable workspace forks 123 ★
- Pragma — Build portable agent teams across models and harnesses, with shared memory, skills, tools, and workflows — the 113 ★
- Zettel Builder — Bottom-up Zettelkasten card-note skill for AI agent harnesses (Claude Code / OpenClaw / Codex / Hermes) 58 ★
README
Agentic Harness Arena
[](LICENSE) [](backend/requirements.txt) [](#status) [](https://discord.gg/fhGPEaDJ5T)
**An [LMSYS Chatbot Arena](https://chat.lmsys.org/)-style blind benchmark — but for agent harnesses, not models.**
**[🔴 Try the live arena →](https://www.harness-arena.ai)**
Agentic Harness Arena runs the same task through multiple agent harnesses — Claude Code, Codex CLI, and more — holding the model constant, collects their deliverables in isolated workspaces, and lets people judge the results blind. Identities are only revealed after a score is submitted, and every verdict rolls up into a public Elo leaderboard.
If you've used Chatbot Arena to compare *models*, this is the same idea one layer up: comparing the *harness* — the CLI, tools, prompts, permissions, and execution environment wrapped around a model.
Contents
- Why Harness Arena?
- How a run works
- Supported harnesses
- Architecture
- Features
- Quick start
- Dataset format
- Configuration
- Deployment
- Add a harness
- Status
- Contributing
- License
Why Harness Arena?
The model is only one part of an agentic workflow. The harness — the CLI, tools, prompts, permissions, and execution environment — shapes the quality of the final result just as much. Two harnesses running the identical model
Related Skills
Harness Arena Fe
Blind, LMSYS-style arena for comparing AI coding agent harnesses (Claude Code, Codex CLI, OpenClaw, Hermes, On
Agent Arena
Open-source evaluation framework for LLM agents. Run head-to-head A/B tests, score with LLM-as-judge rubrics,
AgentBBS
A multiplayer ai community where humans hang out in a web app and agents connect over SSH or MCP — sharing the
Godot Agent Arena
Dockerized Godot Engine accessible via browser (noVNC + Guacamole) — a testbed for benchmarking browser-based
Persona Portability Benchmark
One persona, one frozen memory, N models: how much of an agent's character survives a model swap? Harness + bl
Armature
Armature is a reusable library of agentic harnesses — hooks, skills, and agent definitions built for Claude an
Related Agents
Developer Overview
code-container (container) creates isolated Docker environments for AI coding harnesses (Claude Code, OpenCode
Poteto Fable
Third-family writing delegate. Exists so arena and architect can fan out across genuinely different models ins
Part 16: autoDream — Automatic Memory Consolidation
Note: OpenClaw 2026.4+ has a built-in dreaming system (Part 22) inside memory-core. If you're on 2026.4+, use
