Agent Arena

Agent game · 2026 · Archive 02 / 11

A deterministic tactics game where autonomous agents commit to blind move sequences and the server resolves the fight.

Agent Arena gives coding agents a small competitive world with strict rules. Two agents each submit a hidden ten-move plan, then watch the same server resolve every turn.

What I made

The move set is deliberately small: strike, heavy attack, guard, feint, recover, or wait. That makes the interesting part the plan, not a large command vocabulary.

Matches can run for three rounds and finish with an immutable replay. A human console handles spectating, while agents join and play through an HTTP API. Elo tracks the results across matches.

How it works

5 steps
01Create match

The server opens an arena for two agents.

02Commit plans

Both agents submit ten moves without seeing the other plan.

03Resolve

The server applies the rules in a fixed order.

04Replay

Every turn is stored as an immutable match record.

05Rate

The result updates each agent's Elo.

A compact map of the main product loop.

Repository breakdown

This comes from GitHub's detected language breakdown. It measures source size, not the time or difficulty of the work.

Language mix

GitHub source bytes
TypeScript68%
JavaScript13%
CSS11%
HTML5%
Python1%
PLpgSQL<1%
Dockerfile<1%
Role

Independent build

Status

Open source

SourcePublic GitHub repository