Evaluation Video Wrapper Implementation Plan¶

For Claude: REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.

Goal: Add declarative evaluation-time video recording so Atari and other games can capture reproducible rollout videos directly from the env config.

Architecture: Introduce a dedicated video wrapper resolver in src/rl_training/envs/video.py and wire it into src/rl_training/envs/factory.py. Users opt in via env_kwargs.wrappers.video, and mode-specific config already added in env_kwargs.evaluation keeps video capture isolated to evaluation by default. The wrapper should default to writing under output_dir/videos/<mode> and allow simple trigger configuration without affecting existing training code paths.

Tech Stack: Python 3.10+, Gymnasium RecordVideo, existing env factory/config system, pytest.

Task 1: Add failing video wrapper tests¶

Files: - Modify: tests/test_envs.py

Step 1: Write the failing tests - Add coverage for resolving video wrapper config from wrappers.video. - Add a build-env test proving evaluation mode applies RecordVideo with the expected folder and trigger settings. - Add a build-env test proving the same config does not enable video during training when the wrapper is only present under env_kwargs.evaluation.

Step 2: Run test to verify it fails - Run: pytest -q tests/test_envs.py - Expected: failures because video wrapper resolution and application do not exist yet.

Task 2: Implement video wrapper support¶

Files: - Create: src/rl_training/envs/video.py - Modify: src/rl_training/envs/factory.py - Modify: src/rl_training/envs/__init__.py

Step 1: Write minimal implementation - Add a dataclass for video wrapper settings. - Parse video_folder, name_prefix, episode_trigger_every, step_trigger_every, video_length, fps, and disable_logger. - Apply gym.wrappers.RecordVideo with default output under output_dir/videos/evaluation or output_dir/videos/training.

Step 2: Run focused tests - Run: pytest -q tests/test_envs.py

Task 3: Document video capture usage¶

Files: - Modify: README.md

Step 1: Add docs - Show how to enable evaluation-only video capture with env_kwargs.evaluation.wrappers.video. - Clarify the default output directory and trigger behavior.

Task 4: Verification¶

Run: - Focused: pytest -q tests/test_envs.py - Broader: pytest -q

Notes: - This plan intentionally omits commits because the session instructions forbid committing unless explicitly requested.