We introduce a new benchmark planning domain based on the puzzle game Baba Is You. As in Sokoban, a popular classical planning domain, the agent operates in a fully observable, deterministic grid world and can move in the cardinal directions while pushing objects. However, unlike Sokoban, the properties governing objects in a level are determined by sentences inside the level that the player can manipulate: the state of the world impacts the fundamental dynamics of the environment. This introduces a meta-planning aspect in which the planner must reason not only about object states but also about dynamically reconfigurable rules that result from object locations. Aside from two primitive actions, the behaviour of objects, rules, and movement is modelled using a complex web of derived predicates. Finally, we evaluate the Fast Downward domain independent classical planning system with multiple search configurations on a suite of hand-crafted instances that vary in level size and the number of objects. Our results show that while current planners can solve tiny or sparsely populated levels, their performance degrades sharply as the number of objects or the level size increases. The Baba Is You setting requires thinking outside the box and serves as a highly compelling benchmark for classical planning with axioms. Our results highlight the difficulty of solving classical domains in the presence of complex interacting axioms and the need for improved planners that support them.