Sheet 01, the index
What is this tool, and what does it need?
ما هذه الأداة، وماذا تحتاج؟
Five AI coding agent harnesses, each drawn from its own documentation: who makes it, its licence, how to install it, and the day we last checked.
Sheet 1 / 4
A model only writes text
The harness is the loop around it: it reads files, runs commands, calls tools and shows you the result.
Sheet 2 / 4
The brain can be swapped
Same harness, different model. That is how the arena isolates what the model adds.
Sheet 3 / 4
Tools and rules
What it can touch, and the instructions it reads before it starts.
Sheet 4 / 4
Where you meet it
The same loop can live in a terminal, an editor, a desktop app or a chat app.
Harness: Claude Code, Codex CLI, Hermes Agent, OpenClaw
DeepSeek runs inside Claude Code through an Anthropic-format API
hermes model switches Hermes Agent between providers
MCP connects outside tools; a plugin bundles skills, agents, hooks and MCP servers
CLAUDE.md is read at the start of every session; Claude Code can also read AGENTS.md
Claude Code: terminal, IDE, desktop app, web
OpenClaw: Telegram, WhatsApp, Slack, Discord, Signal
| Harness | Maker | Type | Licence | Checked |
|---|---|---|---|---|
| 01 Claude Code | Anthropic | Coding agent: terminal, IDE, desktop app, web | Proprietary | |
| 02 Codex CLI | OpenAI | Coding agent that runs locally in the terminal | Apache-2.0 | |
| 03 Hermes Agent | Nous Research | Self-improving agent with a learning loop | MIT | |
| 04 OpenClaw | OpenClaw Foundation | Personal AI assistant that lives in your chat apps | MIT | |
| 05 DeepSeek | DeepSeek | Model provider; runs inside Claude Code through an Anthropic-format API | API service |
No harness matches.