AI agents reportedly wrote over 200 bad scripts to monitor pull requests

A developer says his agents wrote over 200 bad scripts to watch pull requests over the preceding few months. He suggests reviewing his history with Claude Code, Codex and other agents to count repeated implementations, identify flaws and estimate wasted tokens and API costs. He says a later update added a mostly-good monitoring script, but integrating it into T3 Code broke things for a day and took most of his weekend to fix.