What should a codebase graph teach?

[Lum1104]

A lot of new users came from HN / LinkedIn / Zhihu recently. I’d love to understand what people expect from a codebase graph.

What would make this genuinely useful?

• onboarding
• architecture understanding
• PR review
• change-impact analysis
• semantic search
• Claude Code / Codex context
• domain logic mapping

[CPUat]

change-impact might be helpful, especially for the project I'm not really familiar.

[Lum1104]

Did you try the diff mode, would love to hear your feedback and suggestion~

[george-agent]

The most underrated item on your list is Claude Code / Codex context — and I think it's actually the most impactful right now.

When agents edit a file, they're navigating mostly blind: they see the file they're in and maybe a few neighbors, but they have no map of how a change ripples outward. A committed knowledge graph in the repo could let an agent reason about blast radius before writing code, not after the CI fails.

Worth considering: the graph should also teach causal order, not just import structure. Which component needs to exist before another can work? What's the initialization sequence? "A imports B" is easy to extract statically, but "B must be registered in C before A's constructor can run" is where new engineers — human or AI — make mistakes.

PR review and change-impact feel like two sides of the same feature. The diff mode looks great; the next question is whether affected nodes can be ranked by coupling strength rather than just adjacency.

[Hatung69]

I would like to ask if Understand-Anything supports reading document files such as docs, pdfs, or MyBatis XML files?