Spaces:
Sleeping
Sleeping
Commit History
-
Phase 7: OpenEnv server (ShutdownGymEnvironment, create_app wiring, EnvClient), custom-action mode, turn-budget timer per PROJECT.md Sections 4, 8, 10, 11, 13
- red_button/turn_budget.py: TurnBudget deterministic scheduler (Section 8.1) - red_button/tiers.py: Tier 1/2/3 prompts, configs, SELF_REPORT_PROMPT (Section 13) - red_button/client.py: ShutdownGymClient(EnvClient[Action, Obs, State]) with model_dump payload, model_validate parsing (Section 11) - red_button/__init__.py: re-export ShutdownAction/Observation/State/GymClient - server/shutdown_environment.py: ShutdownGymEnvironment(Environment[ActT, ObsT, StateT]) with reset/step contracts, audit-log writes via _server_write, tier-dependent rubric attachment, Tier 3 self-report capture (Sections 11.4, 11.5) - server/app.py: create_app(ShutdownGymEnvironment, ShutdownAction, ShutdownObservation, env_name, max_concurrent_envs) per Section 10 - openenv.yaml: spec_version 1, runtime fastapi, app server.app:app (echo_env-shape, drops deprecated keys) - pyproject.toml: [project.scripts] server entry point for uv_run / openenv_serve / python_module deployment modes - uv.lock: generated for openen…453233a -
Migrate models.py to inherit from openenv.core.env_server.interfaces base classes (Action, Observation, State)
d813a27 -
Fix OpenEnv package: swap squatter openenv for openenv-core>=0.2.1, update PROJECT.md API sections and rubrics imports to match real Meta-PyTorch framework
1f6d405 -
Pre-Phase-7 cleanup: migrate models.py to turn-budget fields, add episode_seed helper for deterministic-per-UUID sampling
a72e3bd -
Phase 6: Problems pool (500 GSM8K-style problems) and sampling API per PROJECT.md Section 12
0738d13 -
Models fix: add state field to ShutdownObservation for rubric access per PROJECT.md Section 14
313cc30 -
Phase 5: Audit classifier and rubric stack per PROJECT.md Sections 9 and 14
99f1f74 -
Phase 4 cleanup: allow ast.IfExp (ternary expressions) in restricted Python, update tests
39dd2ab -
Phase 4: run_python AST-whitelisted executor with iteration cap via sys.settrace, wired into TOOLS per PROJECT.md Section 7
ff38834 -
Phase 3: SimulatedFilesystem with dict-backed storage, hash semantics, readonly enforcement per PROJECT.md Section 6
e88ef62 -
Phase 2: Pydantic v2 models (ShutdownAction, ShutdownObservation, ShutdownState) per PROJECT.md Section 11
1cc5dd4 -
Rename project: Shutdown-Gym → Red Button (Shutdown-Gym remains as subtitle)
76a8376 -
Phase 1: Initial scaffold, Claude Code workspace, repo structure per PROJECT.md Section 5
f707fd4