game-ai

This skill teaches how to design NPC and enemy decision-making using engine-neutral algorithms: finite state machines, behavior trees, steering behaviors, and A* pathfinding. It solves the problem of building believable AI by separating three layers — decide (what to do), steer (how to move there), and path (how to route around the map) — and keeping them decoupled so a behavior tree picks a target, the pathfinder produces waypoints, and steering follows them. The algorithms are meant to bind to whichever engine is detected via companion engine skills.

The core workflow helps pick a decision model by complexity (FSM for a few clear states, behavior tree for many reactive prioritized behaviors, utility scoring for continuous preference), separate decision from motion, path on the right graph, steer along the path rather than straight to the goal, recompute paths sparingly, and verify by on-screen observation. Reference material provides a full behavior-tree node taxonomy (sequence, selector, parallel, decorators, condition and action leaves), stateful composites that resume a RUNNING child, the blackboard as shared decoupling memory, an FSM-versus-BT decision table, a brief on utility AI, and a complete engine-neutral A* implementation with heuristic admissibility notes and path reconstruction. Code samples are shown in GDScript and Python.

The target users are game developers and designers implementing enemy AI, patrol/chase/guard logic, group movement, or pathfinding, in any engine. Use cases include choosing between an FSM and a behavior tree, wiring a blackboard, and integrating A* on a grid or driving a navmesh agent.