LLM coding agents live or die on retrieving the right code context, yet how textual code representations affect retrieval and localization has been underexplored. This paper reframes file-level bug localization as a representation-driven retrieval problem and studies which code representations best surface the right files for repository-level tasks. The practical implication: how you encode and index code for retrieval may matter as much as the agent's reasoning when localizing bugs at repo scale.