# predict.sh - save the sandbox's rules as a file, and let a small model read them.
openshell policy get demo --base -o json > policy.json
python3 decide.py policy.json