@TomO, thanks for coming back with an update. Yep, you’re right. Changing environment.json or setup.sh doesn’t always invalidate the snapshot the way it should, especially if the base layer already has a broken git state. That’s a known limitation. A “run agent without snapshot” or “force rebuild” option is on the backlog, but I can’t share an ETA.
Since Aubrey is already in the thread with the team, let’s not duplicate. Please continue there. If it helps speed things up, send her the agent ID and the Request ID from the latest run right top corner of the chat > Copy Request ID, plus the contents of environment.json, so we can pinpoint why the cache bust didn’t work.
As another workaround while we wait for a fix, you can fully recreate the repo connection. Create a new environment on top of the same repo with a different name or a slightly changed structure. That usually gives you a clean snapshot. It’s a workaround, but it works in most cases.
A similar case with a Dockerfile setup is discussed here, in case it helps: How to force the cloud agent environment to rebuild from scratch when using a Dockerfile