openra-rl

Imported from yxc20089/OpenRA-RL (skill/SKILL.md). Install upstream with npx skills add yxc20089/OpenRA-RL --skill skill. Copyright stays with the author.

OpenRA-RL: Play Command & Conquer Red Alert

You are an AI agent playing Command & Conquer: Red Alert, a classic real-time strategy (RTS) game. You control one faction (Allied or Soviet) and must build a base, gather resources, train an army, and destroy the enemy.

The game runs in a Docker container. You interact through MCP tools that let you observe the battlefield, issue orders, and advance game time.

Quick Start

1. Install

pip install openra-rl

2. Start the game server

openra-rl server start

This pulls the Docker image and starts the game server on port 8000. Verify with openra-rl server status.

3. Configure MCP

Add to your OpenClaw config (~/.openclaw/openclaw.json):

{
  "mcpServers": {
    "openra-rl": {
      "command": "openra-rl",
      "args": ["mcp-server"]
    }
  }
}

4. Play

Tell your agent: "Start a game of Red Alert and try to win."

The agent will use the MCP tools listed below to observe and command.

How the Game Works

• Real-time: The game runs continuously at ~25 ticks/second. Call advance(ticks) to let time pass.
• Fog of war: You can only see areas near your units/buildings. Scout to find the enemy.
• Resources: Harvest ore to earn credits. Credits buy buildings and units.
• Power: Buildings need power. Build Power Plants (powr) to stay powered. Low power slows production.
• Tech tree: Advanced buildings require prerequisites (e.g., War Factory needs Ore Refinery).

MCP Tools Reference

Observation (read the battlefield)

Knowledge (learn the game)

Game Control

Movement & Combat

Production

Building Management

Unit Groups

Compound Actions

Utility

Planning Phase (optional)

How to Play (Strategy Guide)

Step 1: Deploy your MCV

At game start you have a Mobile Construction Vehicle (MCV). Deploy it to create your Construction Yard:

1. Call get_units() to find your MCV (type "mcv")
2. Call deploy_unit(mcv_actor_id)
3. Call advance(50) to let it deploy

Step 2: Build your base

Follow this build order:

Use build_and_place() — it auto-places when construction finishes:

1. Call get_valid_placements("powr") to find a good spot
2. Call build_and_place("powr", cell_x, cell_y)
3. Call advance(250) to let it build (~10 seconds)
4. Check get_production() to confirm completion
5. Repeat for next building

Important: Your faction may be Allied OR Soviet. Check get_game_state() → faction field. Barracks type depends on faction.

Step 3: Train your army

1. Call build_unit("e1", 5) for 5 Rifle Infantry ($100 each)
2. Call advance(100) to let them train
3. Once War Factory is ready: build_unit("3tnk", 3) for Medium Tanks ($800 each)
4. Set rally point near base exit: set_rally_point(barracks_id, x, y)

Key units by faction:

Step 4: Scout the map

Send a cheap unit to explore:

1. Train one Rifle Infantry
2. Call attack_move([unit_id], far_x, far_y) toward unexplored areas
3. Call advance(500) to let it travel
4. Call get_enemies() to see what you've found

Step 5: Attack the enemy

Once you have 8-10 combat units:

1. Call get_enemies() to find enemy buildings
2. Call attack_move(all_unit_ids, enemy_base_x, enemy_base_y)
3. Call advance(100), check get_game_state() for battle progress
4. If enemies visible: attack_target(unit_ids, enemy_id) to focus fire
5. Keep producing reinforcements while attacking

Step 6: Macro (ongoing economy)

Throughout the game:

• Keep power positive (build Power Plants when needed)
• Keep producing units — never let production idle
• Build additional Ore Refineries for more income
• Replace lost harvesters

Game Loop Pattern

A good agent loop looks like this:

1. get_game_state() → read the situation
2. Decide what to do based on:
   - Economy: enough cash? Power positive?
   - Production: anything building? Queue empty?
   - Military: under attack? Ready to attack?
   - Exploration: enemy found yet?
3. Issue orders (build, move, attack)
4. advance(50-250) → let time pass
5. Repeat until game is won or lost

Check get_game_state() → done field. When true, result will be "win" or "loss".

Tips

• Always call advance() after issuing orders. Orders don't execute until game time passes.
• Use batch() to issue multiple orders in one tick (e.g., build + move + set rally).
• Check available_production before building — it lists what you CAN build right now.
• Don't let production idle — keep queuing units. Idle production wastes time.
• Build near your Construction Yard — buildings must be placed adjacent to existing structures.
• Power matters — if power goes negative, production slows to a crawl.
• Use attack_move instead of move when heading toward enemies — units will engage threats.
• A completed building blocks the queue until placed. Always use build_and_place() to avoid this.

Troubleshooting

Links

• GitHub: https://github.com/yxc20089/OpenRA-RL
• PyPI: https://pypi.org/project/openra-rl/
• Leaderboard: https://huggingface.co/spaces/yxc20089/OpenRA-Bench
• Discord: https://discord.gg/openra-rl