1 points | by jesusbr 5 hours ago
3 comments
Curious how it handles false positives though. Auto-generated fixes on a misdiagnosed alert sounds like a bad time. How much human review is expected in the loop before a fix gets applied?
Good question — it won’t break prod on a bad diagnosis.
Worst case, you get a draft PR with a bad fix. It only auto-merges if it’s very safe.
There are 5 gates before auto-merge: CI must pass, ≥90% confidence, ≤50 lines changed, self-review ≥70/100, and agent trust level.
If any of those fail, it stays as a draft PR for human review.
Auto-merge is off by default and only unlocked over time as the agent proves itself.
We’d rather open 10 draft PRs than merge 1 bad fix.
False positives are real though — we’re working on improving that with a shared fix database over time.
[dead]
Curious how it handles false positives though. Auto-generated fixes on a misdiagnosed alert sounds like a bad time. How much human review is expected in the loop before a fix gets applied?
Good question — it won’t break prod on a bad diagnosis.
Worst case, you get a draft PR with a bad fix. It only auto-merges if it’s very safe.
There are 5 gates before auto-merge: CI must pass, ≥90% confidence, ≤50 lines changed, self-review ≥70/100, and agent trust level.
If any of those fail, it stays as a draft PR for human review.
Auto-merge is off by default and only unlocked over time as the agent proves itself.
We’d rather open 10 draft PRs than merge 1 bad fix.
False positives are real though — we’re working on improving that with a shared fix database over time.
[dead]