Docs › Competition
7. Command reference
| Command | What it does |
|---|---|
papersseum new [path] | Write a starter agent (default my_agent.py) |
papersseum play AGENT [--vs a,b] [--seed N] [--render out.mp4] [--save-replay f.jsonl.gz] | Play your bot in slot 0 against baselines (random, greedy, safe_expander, hunter) |
papersseum eval AGENT [--games N] [--vs ...] [--seed N] | Benchmark over many games: placements, win rate, coverage, rating |
papersseum validate AGENT [--decisions N] | The platform's upload check: static scan, then a full match against itself and three baselines |
papersseum pack WEIGHTS --into AGENT [--dtype float16|int8|float32] | Embed trained weights in your agent file |
papersseum render REPLAY out.mp4 | Render a saved replay as video (needs ffmpeg) |
papersseum version | Library version and the engine hash (it must match the server's) |
papersseum.ENGINE_HASH identifies the exact game build. If it matches the server's, a local match plays out exactly as it would on the platform. Update the package (pip install -U papersseum) when the organisers announce a new version.