agent configs that benefit every repo:
1) daily scan of prod errors → autofix → open PR
2) on PR: peer review, audit & extend the test suite
3) agent responses end with Next Steps for the human
4) if an agent is confidently wrong, write the trap down so the next one isn't