Suppose we observe something happen in an interaction between two objects A and B. Can we then predict what will happen in an interaction between A and C, or between B and C? Recent research, inspired by work on the 'causal asymmetry', suggests that people use cues to causal agency to guide object-based generalization decisions, even in relatively abstract settings. When object A possesses cues to causal agency (e.g. it moves, remains stable throughout the interaction), people tend to predict that what happened will probably also occur in an interaction between A and C, but not between B and C. Here we replicate and extend this work, with the goal of identifying the cues that people use to determine that an object is a causal agent. In four experiments, we manipulate three properties of the agent and recipient objects. We find that people anchor their inductive generalizations around the agent object when that object possesses all three cues to causal agency, but removing either cue abolishes the asymmetry.