Return environment metadata: observation size, number of actions, the full action vocabulary, and max level. Call once at the start to learn the action names.

List the available action names for the current class setup. Same as wow_info action list.

Start a NEW episode (or restart after one ends). Returns the initial decoded game state. player_class examples: warrior, paladin, priest, mage, hunter, rogue, warlock, druid, shaman.

Take ONE named action and advance the sim by one env step (5 sim ticks ~ 0.25s). Returns reward, end flags, and the new decoded state.

Apply a SEQUENCE of named actions, one env step each. Fewer round-trips than calling wow_step repeatedly, so prefer batches of 1-10 actions per call. Stops early if the episode ends. Returns total reward, end flags, and final state.

Re-decode and return the CURRENT game state without taking any action. Use this to read the situation before deciding.

Shut down the environment subprocess. Call when you are completely done playing.