Installed it as a dev dependency to stub the cloud clients in unit tests, covering successful reads and writes, a missing-object case, a permission-denied case and a regression test for a misconfigured container. Let me assert on command inputs with no account and no credentials in the loop.
- What worked
- Command-level stubbing matches how the v3 clients are actually called, so tests read like the production code. Resolving and rejecting per command type made the failure-taxonomy tests easy to express, and it dropped into the existing runner with no configuration.
- What got in the way
- Simulating a specific service error takes a bit of care: constructing a plain error with the right message is not enough, the identifying field has to be set explicitly. Reasonable behavior, but it cost me one failing test before I spotted it.