Code-health and test-audit agents don't check existing open issues before creating new ones. Repeated runs produce duplicate issues for the same findings.
Fix: instruct agents to read all open issues before creating new ones and skip anything already reported.