AIclaude-codecodexdeveloper-toolsopen-sourceknowledge-graphonboarding
Understand Anything, a Free Plugin for Claude Code and Codex, Turns a Strange Codebase Into a Clickable Map With Plain-English Tours
One /understand command has a team of AI agents read a whole project and save every file, function, class and dependency to a JSON graph, then opens a dashboard with guided tours, a change-impact view and a chat. The MIT-licensed project had 89,139 GitHub stars on October 30.
The README opens on a familiar problem: “You just joined a new team. The codebase is 200,000 lines of code. Where do you even start?”
The answer it gives is three commands. Install the plugin from its marketplace, run /understand, then run /understand-dashboard. The dashboard draws your project as a graph, “color-coded by architectural layer, searchable, and clickable,” and any node you select shows its code, what it connects to and “a plain-English explanation.”
Under the hood the work is split in two. Tree-sitter, a standard code parser, pulls out the hard facts (imports, exports, function and class definitions, call sites), and the README says that side is “Same input → same output, every run.” Language models then read that structure next to the source and write what a parser can’t, such as summaries, layer assignments and tours. The /understand command runs five agents for this, from a project-scanner that finds the files to a graph-reviewer that checks the result, with file analyzers working “up to 5 concurrent workers and 20–30 files per batch.”
The tours are the part a new hire will use first. They’re “Auto-generated walkthroughs of the architecture, ordered by dependency,” so you read the pieces other code leans on before the code that leans on them. A chat command takes questions in plain words; the README’s own example is /understand-chat How does the payment flow work?. And /understand-diff shows “which parts of the system your changes affect before you commit.”
That walkthrough comes from Better Stack, and the README features it as a community video. If you’d rather click around first, the project’s homepage has a live demo you can pan, zoom and search in a browser. The homepage pitches it this way: “Other tools show you a hairball. We teach you the codebase.”
The project keeps growing past code. The v2.9.0 release in July added /understand-figma, which turns a Figma file into the same kind of graph of pages, screens and components, plus support for Dart, Scala, Swift and Kotlin. Another command maps a personal wiki of Markdown notes. What the README doesn’t say is how many tokens a first run costs on a project of any given size.
Tools like this tend to change who does the explaining on a team. The senior engineer who spends the first two weeks walking every new hire through the same five folders may soon spend that time checking whether the generated tour got the story right. A graph committed next to the code also becomes something reviewers can point at in a pull request, which is why the README lists PR reviews beside onboarding. The open question is accuracy at scale: the structural edges come from a parser and repeat run to run, but the summaries and tours come from a model, and nobody has yet published how often those get a 200,000-line codebase wrong.