Using Lima for Coding Agent Isolation
-
Eric McCarthy
-
-
5 min read
With a bit of configuration, I started using Lima to contain Claude Code running in auto mode.
I’m a web-centric software engineer living in Tucson, Arizona. This site, limulus.net, is my personal website.
In March 2018, I got married!
I am most active on fedi, but I also maintain accounts on privately-owned and non-federated platforms:
Note that on Facebook and LinkedIn I only connect with people I know in real life.
I also have some inactive accounts here:
With a bit of configuration, I started using Lima to contain Claude Code running in auto mode.
Lalit Maganti and I have a similar experience with the need to constantly refactor when building a project with a coding agent.
It feels like software quality has been on the decline. It’s understandable to think AI will make it all worse. Yet agentic engineering gives us a real opportunity to do better.
Agents prompted with “red-green TDD” write most of the tests first, then all the code — skipping the part of TDD that actually matters. Dual-loop BDD might be a better prompt strategy.