You can greatly simplify the logic by re-using it from the #129. I've ran your tests with it and found the following differences:

• there's no indirection introduced by injecting CIRRUS_WORKING_DIR/CIRRUS_CLONE_DIR variables to resolve them later — they're resolved in place (since it's already possible)
• in only working (precreated), only working (monorepo), both (precreated) and both (monorepo + precreated) tests the user-provided CIRRUS_WORKING_DIR/CIRRUS_CLONE_DIR variables are respected even if the pre-created directory exists, which is how it works right now
• in only clone and only clone (precreated) tests a scratch working directory is provided in the absence CIRRUS_WORKING_DIR, thus making possible to use the configuration pattern described in Separate CIRRUS_CLONE_DIR from CIRRUS_WORKING_DIR #106

• in only clone and only clone (precreated) tests a scratch working directory is provided in the absence CIRRUS_WORKING_DIR, thus making possible to use the configuration pattern described in Separate CIRRUS_CLONE_DIR from CIRRUS_WORKING_DIR #106

Err: this allows a user to both clone into some non-standard directory and run some script's in an empty working directory, which may make sense in some cases.