When should you use pytest skip, xfail, or quarantine a test?

The condition should be narrow and evaluated from an actual capability, not a broad guess such as skipping all CI runs. For xfail I decide whether the failure should occur during execution and whether unexpected passing should fail the build. I avoid catching arbitrary exceptions inside the test because that can convert a new regression into the expected result. Reports must make skipped, expected-failed, and quarantined counts easy to see.