Triage every production alert
Point Sentry at one URL. Every alert that arrives is checked against what you already triaged, rated against your own definition of critical, and written up — you are interrupted only when it earns it.
Run by CTO.
What you fill in
- What does this service do, and who is on the other end?
- Written once, read on every alert. This is what turns 'TypeError at line 42' into 'nobody can pay'.
- What counts as critical for this business?
- A payment failure and a thumbnail failure look identical to a monitoring tool. Only you know which one costs money.
- Which alerts will you point at this URL?
- Filter this in the alert rule, not here: the webhook accepts 10 fires per minute per trigger, and one bad deploy can produce fifty identical alerts. Every fire past that limit is recorded as failed and never runs.
- What should happen for each alert?
- Tick what you'd do yourself at 3am. Everything unticked is work that does not happen.
- What is worth interrupting you for?
- Everything under this line waits in the report until you go and look.
- When the same error fires again
- One incident produces the same alert dozens of times. Logging the sighting is what keeps the fiftieth from reading like the first.
- Where the ticket goes
- Linear, Jira or GitHub Issues if one is connected. Left empty, it appends to shared/incidents/backlog.md.
- Never, under any circumstances
- It already deploys nothing, restarts nothing and closes no alert. Write what else is off limits — this one runs unwatched, in your name, at 3am.
What happens when you start it
- Read the alert
- Check whether it is already known
- Measure what it breaks
- Rank the likely causes
- Report and raise what matters
- Sentry — optional
- GitHub — optional
- Linear|Jira — optional
What does "Triage every production alert" do?
You fill in 8 fields — what does this service do, and who is on the other end, what counts as critical for this business, which alerts will you point at this url, what should happen for each alert, what is worth interrupting you for, when the same error fires again, where the ticket goes and never, under any circumstances — and an AI CTO takes it from there: read the alert, check whether it is already known, measure what it breaks, rank the likely causes, report and raise what matters. You can close the tab — the run continues on the server, and what lands is a finished file in your Content, not a conversation to read back.
Who runs it, and how long does it take?
The AI CTO — hire one and this comes with it, with nothing to install or switch on. A run takes about 2 minutes and you are told when the file lands.
What do I need to connect first?
It works better with Sentry, GitHub, Linear or Jira.
All OpenLabor superpowers