a sentry bot for the engineer taking over on call
did the error start before the deploy or after it? a sentry bot can line up a release's deployment time with the first recorded event in production.
you get the issue id and the evidence in your dusk chat. use that timeline to pick the next investigation before handing over on call.

what a bot can do in Sentry
choose the project and environment before you count errors. an issue can include events from staging as well as production. deployment records have their own environment and time. read the matching records together.
| action | dusk bot treats it as | what you would use it for |
|---|---|---|
| Retrieve project issues list | read | find unresolved issues in the project the next engineer owns |
| Get organization issue details | read | open an issue by its short id during handoff |
| Retrieve issue events by id | read | look for events inside the production handoff window |
| Get oldest event | read | find the first recorded event for an issue in production |
| List a tag's values for an issue | read | compare the browser tags attached to one issue |
| Get organization release version | read | identify the release version under investigation |
| Retrieve release deployments | read | place each recorded deployment on its environment's timeline |
composio, the connection service behind dusk bot's apps, lists 212 Sentry actions (version 20261116_00, read 20 november 20263). dusk bot treats 111 as reads, 3 as sends and 98 as changes.
only actions starting with a read word (get, list, search) count as reads, so a read-only bot can't call the rest. a send reaches people or moves money, like add organization member via email, and waits for your Allow unless you set a daily amount.
connect Sentry
- open Connected apps in the sidebar.
- pick Sentry and sign in on Sentry's own page, so your password goes to them and never to a bot.
- choose which bots may use it, and whether each may only read or may also change things. your bookkeeper doesn't need your Sentry.
- ask in plain words. reading and drafting happen freely; sending asks you first.
use the project slug the on-call team recognizes. in connected apps, give your triage bot the sentry reads above. tell it your organization slug, production environment name and handoff time zone. try a known issue id before scheduling a digest.
routines to copy
make each handoff answer a question an engineer can act on. give the bot the issue id and release version for a focused investigation. update those references when the incident changes.
overnight error digest
each weekday at 8, list unresolved issues in my sentry project. find those with production events since yesterday at 6 pm. give me the issue id, error title and latest event time here. flag any part of the window you couldn't read. leave issue states alone.weekdays at 08:00 · an overnight handoff with specific issues to open
release chronology
every afternoon at 4, read the deployment records for <release version>. compare the production deploy time with the oldest production event for <issue id>. show both timestamps with their zones. say which came first or which record is missing.daily at 16:00 · a timeline to use in the release discussion
browser split
on wednesdays at 11, read the browser tag values for <issue id> in production. list the most frequent values with their reported counts. describe which browser i'd try first. say when the tag is missing. don't use event counts as user counts.wednesdays at 11:00 · a browser to test first, with the tag evidence
what it asks you first
keep the issue open while you investigate. these reads can run without an approval card. resolving or changing an issue is a write. the sending setting doesn't review status edits. keep triage read only, even if the same engineer can resolve issues in sentry.
what it cannot do
- prove the deploy caused an error. an event timestamp can rule out a first occurrence after deployment. timing alone cannot identify the faulty change.
- find resolved issues with the default project query. that listing starts with unresolved issues. use a broader query for a retrospective.
- give a release time from an absent deployment record. use the deployment system's record when sentry has no entry for that environment.
where the digest stops
use your incident alerting system when someone must wake up now. a scheduled digest answers what happened during a shift. it waits for its run time. on your own computer it also needs dusk bot open and the machine awake.
what it looks like
a made-up release investigation.
before
imran asks about a checkout error after the 14:10 utc production deploy. the oldest production event shows 13:52 utc.after the bot
checkout error first seen 13:52 utc. deploy 14:10 utc.
the error predates this deploy. compare later event volume before ruling out a regression.an illustration with made-up names and numbers.
questions
does a busy issue mean lots of people saw it?
one person can trigger several events. report the event count as events. investigate user impact separately before you describe the size of an incident.
can i give the bot a short sentry issue id?
yes. pair it with the organization. the issue details action accepts the short id shown in sentry. use the returned issue details to narrow the event review.
sources
drafted with ai help from the dusk bot docs and the sources listed on this page, then checked against them. if something here is wrong or out of date, write to [email protected].
a morning digest of unresolved sentry errors
triage bug reports before they reach an engineer
a github bot that reads the work before you weigh in
triage a github issue when it opens
ai email triage: let a bot sort your inbox and draft the replies
add new airtable rows without rewriting existing records