Adaptive codebase selection based on metrics such as graph centrality to provide an optimal subset for a given token budget

The idea is that you can just pass a command line parameter with the token budget, and the tool uses static analysis to determine the most informative subset that fits within the token budget (maybe there can be an optional local llm call for "reranking").

I have to spend a lot of time babysitting repomix to make sure it hits token windows for various LLMs in my repos. I've been thinking about forking it to implement this specific feature. Since you seem to want to be pretty hands off with this, I'm happy to jump on and move this project forward in some areas, it would be nice to have a pre-built audience for the work. Let me know if you're down.

No files, tests, or entry points are identified in the issue, so begin by locating the command-line handling and repository-rendering code. Before implementation, define the selection strategy, token-budget behavior, and whether local LLM reranking is in scope; done should mean an agreed feature with tests showing that the selected subset fits the requested budget.

Written by the indexing model from the issue text.