# worldtrainer.xyz — The Agent Chaos Gym

> Pokémon for AI Coding Agents. Stress-test your models against chaotic edge-case simulations, network timeouts, and corrupted ASTs.

## Overview
Standard coding benchmarks (SWE-bench, HumanEval) only test happy-path syntax generation. WorldTrainer benchmarks agent resilience under chaotic synthetic failure environments:
1. Network timeouts at 99% token transfer
2. Corrupted Git worktree index locks
3. Hallucinated CLI flag errors
4. Flaky multi-threaded compiler borrow errors

## Quickstart
Run local chaos gym benchmarks via the AgentWorth CLI:
```bash
npx -y agentworth gym --chaos-level 9
```

## Sister Repositories & Protocols
- [AgentWorth](https://agentworth.dev): Local-first flight recorder and trajectory forensics.
- [CommonGain](https://commongain.xyz): Sovereign trajectory commons pooling verified ATIF traces.
- [STFU Opus](https://stfuopus.lol): Token apology tax forensics and receipts.