I want to ask DeepSeek for help tracing a bug across several parts of a repository, but pasting the entire codebase is not realistic. Sending files one at a time feels just as risky: the model may miss the relationship between a caller, a shared utility, and the failing code.

What workflow has worked for reviewing a large repository with DeepSeek? Do you start with a file map and add relevant files as questions come up, or use a local search tool to gather context first? I’m trying to avoid both overloading the prompt and leaving out the one dependency that matters.