We just reworked the diagram generation logic using tree-sitter AST parsing. It's much faster and should not exceed context size of the best models when handling bigger repositories. Take a look and let me know if it works for you!
Short answer: not yet. The tool may recognize dynamic import patterns in your code, but it doesn't resolve what actually gets loaded at runtime since there's no execution tracking (yet). Happy to take a PR or issue if this is something you'd find useful!
This is a really cool project, and incredible traction - great job! I can see how this would work well with Claude Code to quickly explain what's going on.
I've worked on a similar project, but instead of going the ASCII route, we decided to lean into interactivity for code exploration. The angle is more around empowering human understanding, instead of agent understanding. We also chose D2 (and are looking to support Mermaid soon too).
Would love to hear others' thoughts on whether interactivity for humans is more valuable vs context for agents.