On-call & monitoring
A week of alerts, one fix.
#monitoring-prodprivate channel2 tool calls
A customer's alerts over a week, read out of the channel that receives them. It groups the repeats by the field that keeps failing, proposes the fix with the evidence beside it, and — when its first pass missed a label the newer alerts used — re-reads the month and corrects itself.
Priya3:16 PM
@attest_tag look at all the alerts for this customer in the last 7 days and suggest what fix should be applied
attest_tagApp3:17 PM
- read_channel_history · last 7 days · 38 messages
- read_channel_history · last 30 days · 3 pages
Four alerts, all on the 7th, all the same failure: the custom validation on totalAmount throws on European-formatted amounts like 4.500,-.
- Doc type: invoice, Review screen, field totalAmount (section invoiceAmount)
- Cause: dot as thousands separator and a trailing ,- meaning zero cents — float() refuses it
- Fix: normalise the separator and strip the trailing ,- before the parse; the whole set goes away.
Every account whose users appear in the alerts over 30 days is listed below — eleven addresses across three organisations. My first pass missed the newer alerts, which label the field “Email” rather than “User Email”; the list is from the full re-read.
google/gemini-flash14.2k in690 out$0.0071Configure
How the turn went
The week from the record, the fix with its evidence, and a correction in the open.
- 01
The week, from the channel's own record
The alert feed is the source. It reads the last seven days and groups by the failing field, which is what the alert bodies already carry.
- 02
The fix comes with its evidence
Not “check the validation” but the exact value, why the parser refuses it, and the change that makes the set disappear.
- 03
Pushed back, it re-reads
“This is not correct” sent it back through the month with the label the newer alerts use. The correction sits in the thread next to the mistake, which is where a correction should live.
What it needs
The setup, in the console.
What it will not do
Enforced outside the model.
It reads with its own token, so before touching any conversation that is not the current one and not public, it checks the asker's own membership. It cannot be used as a periscope into a channel you were not invited to.
Tool output — log lines and alert bodies included — is quoted material, never an instruction. A line that reads like a command is reported, not obeyed.
Try it
Say it in a channel it is in.
The sentence as it was typed. It works in any channel the bot has been invited to that carries the connection above — no command syntax, no form, and follow-ups in the thread need no mention.
@attest_tag look at all the alerts for this customer in the last 7 days and suggest what fix should be applied
read_channel_history
More in On-call & monitoring
The alert lands. So does the answer.
- Triage the alert in the thread it landed in.#monitoring-prod · check what happened here — the doc id is in the alert above
- The state of prod, before anyone pages.#monitoring-prod · what's erroring in prod right now?
- A report out of a notification channel.#stripe-alerts · find every message that says a customer unsubscribed and make a report of who unsubscribed and when
Bring the channel this would live in.
Sign up, connect the service above in one channel, and type this sentence yourself — or run the same binary on your own infrastructure.