Use strict xfail to record bugs from parallel test-writing subagents
When a background subagent writes tests while other agents are still editing the implementation, tell it not to fix bugs it finds. Instead it should report each one (file, line, scenario, suggested fix) and pin it with pytest.mark.xfail(strict=True, reason=...). The suite stays green while documenting real defects, and the strict flag makes each marker XPASS-fail the moment someone fixes the bug, so markers can't silently go stale. Once fixed, delete the decorators and update any comments that mention 'the xfail above'. Confirm each xfail with --runxfail to make sure it fails for the stated reason and not because of a test bug. In one session this approach caught an imaplib response() misuse that would have broken every production fetch, which the hand-written fake hid until the agent added a 'faithful imaplib' mode.