Datasets:
Download tools/harbor/Dockerfile from THUSI-Lab/Video-Arena: direct link, hf CLI and curl.
- Browser
- Download file 2.18 kB
-
https://huggingface.co/datasets/THUSI-Lab/Video-Arena/resolve/main/tools/harbor/Dockerfile
- Command line
-
hf download hf://datasets/THUSI-Lab/Video-Arena/tools/harbor/Dockerfile
-
curl -L -o Dockerfile https://huggingface.co/datasets/THUSI-Lab/Video-Arena/resolve/main/tools/harbor/Dockerfile
2.18 kB
| # Video-Arena sandbox for Harbor (Protocol B): offline, with the tools pre-installed. | |
| # The same image is the verifier: scoring needs only python3 and the case's judge (stdlib). | |
| # | |
| # docker build -t video-arena/base:latest tools/harbor | |
| # | |
| # Everything the agent may use is installed here, because the sandbox has no network at run | |
| # time: ffmpeg/ffprobe to read the replay, Python with the usual video and numeric stack, and | |
| # tesseract for on-screen text. Deliberately absent: game engines, engine weights, ground truth, | |
| # and model weights of any kind. | |
| FROM python:3.11-slim-bookworm | |
| ENV DEBIAN_FRONTEND=noninteractive \ | |
| PIP_NO_CACHE_DIR=1 \ | |
| PIP_DISABLE_PIP_VERSION_CHECK=1 \ | |
| PYTHONDONTWRITEBYTECODE=1 \ | |
| MPLBACKEND=Agg | |
| # tmux (and asciinema) are what Harbor's terminal agents such as terminus-2 drive the shell | |
| # through; installed here because they cannot be fetched once the trial is offline. | |
| # libgl1/libglib2.0-0 are the shared objects opencv links against. | |
| RUN apt-get update \ | |
| && apt-get install -y --no-install-recommends \ | |
| ffmpeg tesseract-ocr tesseract-ocr-eng \ | |
| tmux asciinema \ | |
| jq bc file procps less vim-tiny \ | |
| libgl1 libglib2.0-0 \ | |
| && rm -rf /var/lib/apt/lists/* | |
| RUN pip install \ | |
| numpy scipy pandas \ | |
| pillow opencv-python-headless scikit-image imageio imageio-ffmpeg av \ | |
| matplotlib pytesseract tqdm | |
| # /app is the case folder. The task mounts match.mp4 and instruction.md into it read-only; | |
| # answer.json is the pre-seeded empty answer (the schema's root key, an empty list), which the | |
| # agent overwrites in place. Every game uses the same root key, so one seed serves all 180. | |
| WORKDIR /app | |
| RUN printf '{"events": []}\n' > /app/answer.json \ | |
| && chmod 666 /app/answer.json \ | |
| && mkdir -p /logs/verifier /logs/agent /logs/artifacts | |
| # The unprivileged user the agent runs as ([agent].user in every task.toml): agent harnesses | |
| # such as Claude Code misbehave as root. It owns /app (scratch space next to the read-only | |
| # inputs, as in the native sandbox); the verifier stays root. | |
| RUN useradd -m -u 1000 -s /bin/bash agent && chown agent /app | |
| CMD ["sh", "-c", "sleep infinity"] | |