dependency-cruiser: a computational sensor for your AI agents

Do you know dependency-cruiser?

It is a tool that reads your codebase and maps who imports what.

You write rules: “the domain never imports the infrastructure”, “no cycles”, “the UI does not touch the DB”.

It runs in CI and warns as soon as an import line violates a rule.

A computational sensor for AI

Many people see it as an architecture tool, but I see it as a computational sensor for AI.

Because your agent (Claude Code, Cursor, whichever) will take shortcuts. Always. A direct import of the DB from a component, “it works”.

With dependency-cruiser in the harness, the shortcut blows up in CI and the agent has to redo it properly. Without you having to re-read 40 files.

The twist

It only works if you have a clear architecture to describe. Sharp boundaries, strict rules.

On a codebase where everything imports everything, there is nothing to check. The sensor sees nothing. No rule is possible.

And that is where it gets vicious: the more chaotic the codebase, the fewer guardrails the AI has, the more chaos it adds.

The AI gets worse and worse on your project. Not because the model is degrading. Because the terrain is.

So before stacking agents on an app generated in 3 weeks, have someone who does this for a living look at the structure. That is what decides whether AI is still helping you in 6 months… or not at all.

PS: dependency-cruiser is free, open source, and installs in 10 minutes.