An L2 support engineer that's just Claude Code on a /loop, with all its state living in Slack reactions. No app, no database, no deploy — and it only pings you when it's real.
Key takeaways
Use Slack reactions as the state machine: 👀 means claimed, ✅ means done, so overlapping five-minute passes skip claimed alerts and never double-file.
Dedupe against closed issues as well as open ones, or every won't-fix you close comes back the next time the error fires.
Have the bot ping only when the real PR gates pass, not the chronically-red nightly job, so every DM is a signal you can trust.
