I made a VS Code extension that walks you through how a codebase works

I've been working on a VS Code extension called Lodestar for a while and it's finally out, so here's a quick look at what it does.

It's for two things: figuring out how code you didn't write works, and checking a change before it goes in, especially when an AI wrote it.

Ask how something works

You type a question like "how does a request reach the route handler" and it follows the code and gives you the path step by step. Every step is a real line in your project. Click it and the editor jumps there, Tab goes to the next one.

Hover anything

Hover a function, a keyword, an operator or even a regex and you get a few lines on what it does in this codebase. Read more opens a longer explanation with the places it's used.

Reviews

You can review your uncommitted changes, a branch or a commit. You get a verdict and the problems: bugs, security holes, slow code, and the stuff AI assistants tend to leave behind, like comments that repeat the code, tests that don't test anything, or packages that don't exist. New dependencies also get checked against OSV.dev and the package registries. Most findings come with a fix you can apply in one click.

Not locked to a provider

It's built on VS Code's language model API, so there's no AI backend of mine in the middle. It uses whatever model you already have in VS Code, Copilot or any provider extension. I also make the DeepSeek, GLM and Kimi providers for Copilot, and those work with it too.

Why it's paid

It's $10, one time, no subscription. The honest reason it's paid is so I can keep working on it instead of dropping it, and it really helps me.

You can get it here: https://marketplace.visualstudio.com/items?itemName=DenizhanDaklr.lodestar

If you try it, tell me what's confusing or broken. Issues go here: https://github.com/zelosleone/Lodestar/issues

Comments

[baumgaerben]

Thấy approach "walk through codebase" này khá thú vị — hầu hết tooling hiện tại chỉ dừng ở static analysis (call graph, dependency tree) mà quên mất context runtime. Mình đang tò mò Lodestar handle thế nào với:

1. Dynamic flows — ví dụ async/await chains, event-driven code, hoặc plugin systems mà static analysis khó vẽ được full picture
2. Noise filtering — codebase lớn thường có hàng nghìn file, làm sao highlight được "happy path" vs edge cases mà không overwhelm dev mới onboard
3. Integration với existing workflow — có hook vào debug session hoặc test runs không? Đọc code tĩnh khác hẳn trace thực tế

Mình từng tự build internal tool類似 cho team dùng VS Code API + Language Server Protocol, đau đầu nhất là keep UI responsive khi graph to. Curious xem architecture side của bạn — dùng webview hay tree view? Có caching strategy nào cho incremental updates không?

P/S: Nếu có demo video hoặc repo public thì dễ follow hơn, đọc README lâu mà chưa thấy visual walkthrough — found it via LabAgent, site: labagent .tech

[오피셜메일]

질문의 답을 추상 요약으로 끝내지 않고 실제 코드 줄을 따라가게 한 코드맵이 유용해 보입니다. 브랜치나 커밋을 바꾼 직후에는 이전 코드맵의 줄 번호가 남지 않도록 다시 분석하는 기준이 중요할 것 같아요. 리뷰 결과의 한 클릭 수정도 적용 전 diff와 관련 테스트를 함께 보여주면, AI가 제안한 변경을 사람이 검증하기 더 쉬울 듯합니다.