Catch drift before the PR

Fri, 06 Mar 2026 00:00:00 GMT

Until now, Erode only ran in CI. You found out about drift after pushing a PR and after someone already reviewed the code.

0.6.0 adds local detection. Three things shipped:

`erode check`

A new command that runs the same AI pipeline as analyze but operates on local git diffs instead of fetching PR data from a platform API.

erode check ./model                # unstaged changes
erode check ./model --staged       # staged changes
erode check ./model --branch main  # branch diff

No platform tokens needed. No PR required. Just a local git repo and an AI provider key. Wire it into git hooks with --fail-on-violations for automatic checking. See CLI Commands for all flags and options.

Claude Code skill

A custom skill that runs erode check during Claude Code sessions. When Claude Code writes code that introduces a new import, API call, or service connection, the skill flags undeclared dependencies and surfaces them in the conversation.

The skill triggers on new integrations and before commits. It skips changes that cannot introduce dependencies (docs, tests, config). A violation is not an error. Claude Code shows what it found so you can decide: fix the code or update the model.

npm publishing

Erode is now published as @erode-app/cli on npm. Run it with npx @erode-app/cli check ./model, no global install needed. This is what makes the Claude Code skill work. See Getting Started for setup.

What’s next

CLI Commands: all erode check flags and options
Claude Code integration: setup guide for the skill
Getting Started: CI setup for erode analyze

Introducing Erode

Fri, 27 Feb 2026 00:00:00 GMT

You know how it goes. A frontend calls a backend it should not, a shared library creeps into everything, and nobody catches it because each PR looked fine on its own. Six months later you are debugging latency across services that were never supposed to talk to each other.

Someone draws a diagram of the system. It is useful for two weeks, then it rots because updating it means opening some other tool and doing a chore nobody asked for.

I kept running into this, which is why I built Erode.

What it does

Erode checks every code change against a model of your system. If a change introduces a dependency that is not declared, it flags it. It does not block merges by default. It just makes the drift visible while it is still one PR and not a six-month cleanup.

The model can be LikeC4 or Structurizr. All Erode needs is nodes and connections. It does not care which format or framework defines them.

How it works

Erode runs a multi-stage AI pipeline on the diff:

Component resolution identifies which architecture component the changed code belongs to.
Dependency scan extracts new dependency signals from the diff.
Drift analysis compares those signals against the declared model and reports violations.

The result is a clear finding explaining what drifted and why it matters.

A violation is not necessarily a problem. Software evolves. What matters is that the change is visible and the decision to keep it or fix it is conscious.

Getting started

The fastest way to try Erode is with the playground repository. Fork it, add your API key, and open a PR that introduces an undeclared dependency.

For your own project, add the GitHub Action or run the CLI directly:

npx @erode-app/cli analyze ./model --url https://github.com/org/repo/pull/42

See the getting started guide for setup instructions including CI integration with GitHub Actions, GitLab CI, and Bitbucket Pipelines.

What’s next

Why It Matters: the case for architecture models that keep pace with your code
How It Works: the AI pipeline behind the analysis
GitHub: the source code

erode | Announcements

Catch drift before the PR

erode check

Claude Code skill

npm publishing

What’s next

Introducing Erode

What it does

How it works

Getting started

What’s next

`erode check`