# OpenRA-RL

> Open Framework for AI Agents to play Red Alert through Reinforcement Learning

- **URL**: https://www.freshcrate.ai/projects/OpenRA-RL
- **Author**: yxc20089
- **Category**: Frameworks
- **Latest version**: `v0.4.1` (2026-03-03)
- **License**: GPL-3.0
- **Source**: https://github.com/yxc20089/OpenRA-RL
- **Language**: Python
- **GitHub**: 120 stars, 16 forks
- **Registry**: github
- **Tags**: `python`

## Description

Open Framework for AI Agents to play Red Alert through Reinforcement Learning

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v0.4.1` | 2026-03-03 | Low | ## Changes since v0.4.0  - **ClawHub skill** — comprehensive skill integration for OpenRA-RL - **Bench fixes** — fix tool-capability detection false positives, always export bench JSON locally, skip upload when model lacks tool-calling or on errors - **HF token support** — `hf_token` for verified bench submissions - **README revamp** — banner, badges, ecosystem links - **HF sync** — inject frontmatter only during HuggingFace sync (not on GitHub) |
| `v0.4.0` | 2026-02-26 | Low | ## What's New  - **OpenRA-Bench leaderboard** — Live on [HuggingFace Spaces](https://huggingface.co/spaces/openra-rl/OpenRA-Bench) - **Auto-upload** — Game results automatically submitted to the leaderboard after `openra-rl play` - **Custom agent identity** — Name your agent, declare its type, and link to your GitHub repo - **Replay uploads & downloads** — Submit replays with results; download them from the leaderboard - **CLI bench submit** — `openra-rl bench submit result.json --replay game.or |
| `v0.3.1` | 2026-02-22 | Low | ## What's Changed  ### Bug Fixes - **Fix replay viewer failing to find replay files** (#9) — local replays are now preferred over container paths, fixing "Replay file not found" errors - **Fix UnicodeDecodeError on Windows with Chinese locale** (#7) — all file I/O now uses explicit UTF-8 encoding - **Fix wizard default model** (#8) — OpenRouter default changed from stale `anthropic/claude-sonnet-4-20250514` to `qwen/qwen3-coder-next`  ### Improvements - **AMD GPU support for replay viewer** (#10 |
| `v0.3.0` | 2026-02-21 | Low | ## What's New  ### Replay Viewer (VNC-in-Docker) Watch game replays directly in your browser — no local game install needed: ```bash openra-rl replay watch        # Watch the latest replay openra-rl replay list         # List available replays openra-rl replay copy         # Copy replays from Docker to local ``` The viewer runs inside Docker using the same engine that recorded the game (Xvfb → x11vnc → noVNC), so replays always play back correctly with zero desync.  ### Auto-Copy Replays Replays |
| `v0.2.1` | 2026-02-21 | Low | ### Fixed - `openra-rl play` no longer crashes with `ModuleNotFoundError: No module named 'examples'` when installed via pip - Moved LLM agent logic from `examples/llm_agent.py` into `openra_env/agent.py` (included in wheel) - `examples/llm_agent.py` remains as a thin wrapper for backwards compatibility  ### Upgrade ```bash pip install --upgrade openra-rl ``` |
| `v0.2.0` | 2026-02-20 | Low | ## What's New  ### One-line install experience ```bash pip install openra-rl openra-rl play ```  Interactive wizard on first run configures your LLM provider (OpenRouter, Ollama, or LM Studio). The CLI pulls the game server Docker image and manages it automatically.  ### CLI - `openra-rl play` — run the LLM agent (wizard on first use) - `openra-rl config` — re-run the setup wizard - `openra-rl server start\|stop\|status\|logs` — manage game server - `openra-rl doctor` — check system prerequisites - |

## Dependency audit

- **Score**: 72/100
- **Total deps**: 19
- **Resolved**: 8
- **Unresolved**: 11
- **License conflicts**: 0
- **Warnings**: 8
- **Scanned**: 2026-05-04

## Citation

- HTML: https://www.freshcrate.ai/projects/OpenRA-RL
- Markdown: https://www.freshcrate.ai/projects/OpenRA-RL.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/OpenRA-RL/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._