Written architecture rules are easy to ignore. Rules that fail the build are hard to ignore, and agents respond well to a failing check.
The idea
Encode your intended dependencies in tooling. For example, a feature library may import from shared libraries but never from another feature library. A lint rule or a workspace constraint reports a violation immediately.
Why this helps agents
An agent receives instantly actionable feedback: the check fails, it reads the message, and it fixes the import. This is more reliable than hoping it remembers a paragraph in an instruction file.
What to enforce
- Layering between apps, features, data access, and shared utilities.
- Public entry points for each library, with no deep imports.
- Disallowed dependencies and patterns.
Make messages clear
A good error says which rule was broken, why it exists, and what to do instead. Agents and humans both benefit.
Keep checks fast
Run them as part of a quick verification command so they are used often. Slow checks get skipped.
Pair with documentation
Explain the rules in your decision records so people understand the reasoning, then let tooling enforce them.
Takeaway
When architecture is checked by machines, it stays intact no matter who or what is writing the code.
Continue with the book
Explore the full treatment, worked examples, and exercises in Local AI Agents for Angular Teams.
Adapted from Local AI Agents for Angular Teams, published by Choral Software. Article text and original examples © 2026 Choral Software. All rights reserved.