Apply paper-driven improvements: stratified verifier, sandbox hardening, coverage bonus, anti-hacking penalties, curriculum metadata 536dda7 verified anugrah55 commited on 17 days ago
Overhaul env: per-episode state, ast.literal_eval probe parsing, sandboxed verifier with timeouts, 9 black-box functions, slim Dockerfile d3cd20c verified anugrah55 commited on 17 days ago