dumped every game idea i've ever had, shipped, prototyped, or sketched into the wiki as its own page, and asked the agent to read the whole corpus and tell me my style. the synthesis is almost embarrassingly clean. they are all one verb and a twist: deliver, probe, rummage, watch and match, sniff and accuse. i never design a menu of verbs, and i never build exotic maps. the setting is always a place i actually know, an apartment, a village, a bus, a parking lot, a campus, a church, a printer in my own house, and the game is corrupting that familiar space until it betrays you.
the motif is consistent enough to name: the mundane space that turns on you. the apartment thing, the village that inverts, the printer head that exits reality, the tracker that's inside your own arm, the traffic camera feed that converges on your own lot. half the ideas make work itself the monster, the gig delivery, the janitor job, the busy work, the office. and the ending is always the product, a point of view swap, a knock on the mirror then black, the stars going out. i write for the last thirty seconds, which is exactly what makes games clip friendly, plus the titles are comedy and the games are horror, and that tension is the brand.
the data pulled back the same pattern from a different direction. my games that are first person, one verb, a space you know, and the grungy retro skin are the ones that convert and grow, while the clean pixel puzzles and sci fi exploration underperform. the recommendation that fell out is concrete: finish and ship the traffic camera game and one of the finished horror prototypes, and the deep sea job horror concept that's still just a doc is the highest leverage idea on the board because it's the purest version of the motif. the winning move is lane discipline, not more invention.