Find What Caused a Sentry Spike
For the error graph that just went vertical. DuckyHelper reads the issue in Sentry, finds the deploy that landed right before it, points to the line that changed, and drafts a note for the team with the rollback command.
“Find what caused this Sentry spike: match it to a deploy, show the changed line, and draft a heads-up with the rollback.”
- 01Learn the web APIs of Sentry and GitHub, first time onlyBuilt from each site's own buttons, forms and requests. Kept on your device and reused next time.
getIssue(id)getEvents(issue)listDeployments(repo)compareCommits(base, head) - 02Read the error, the stack and who's affected
- 03Find the deploy right before the spike
- 04Point to the line that changed
- 05Draft a heads-up with the rollback command
You getThe cause, fast. You know which change did it and how to undo it. The next spike starts with every site already learned.
A run, drawn from this page · on Mac
Nothing to send, so it runs straight through.
Nothing in this job sends a message or pays for anything, so it works through to the end and tells you when it's done. It does not stop to ask before it acts, so keep an eye on any task that touches messages or money.
If it hits a sign-in, you sign in yourself. It's built not to type your passwords or card numbers.
Do it once. Repeats come later.
Repeating it on its own (whenever a new one comes in) is not in the app yet. Today you ask again when you need it, and each run goes through the API it learned the first time, so it's quick.
Before you ask it.
What do I say?
Press fn (Ctrl+Shift+Space on Windows and Linux) and ask out loud, in your own words, like "Find what caused this Sentry spike: match it to a deploy, show the changed line, and draft a heads-up with the rollback."
Why is the second time faster?
The first time, it writes small JavaScript tools for Sentry, GitHub and Slack and builds an API from each site's own buttons, forms and requests. After that it calls them directly, so the job runs in one go instead of click by click. It costs less and breaks less.
Will it send anything without me?
It doesn't stop to ask for now: when the job sends a message or pays, it does it. A Stop control stays on the window the whole time, Take control hands it back to you, and every task is recorded so you can watch it again. Keep an eye on tasks that involve messages or money. It's built not to type your passwords or card numbers: you sign in yourself, and it picks up from there.
Why not just do it by hand?
Spikes start at a deploy, but finding which change did it means reading diffs under pressure.
Can it keep doing this on its own?
Not yet. Repeating a job on its own (whenever a new one comes in) is not in the app yet. Today you ask again when you need it, in the Mac app.
Can I use this today?
Yes: doing jobs in your apps is in the Mac app, for everyone. Talking to DuckyHelper, and having it see your screen and draw on it, works on Mac, Windows and Linux.
- On Mac
“Deploy the latest commit on main to production and tell me when it's live.”
Ship the Latest Commit and Watch the Build
See it run - On Mac
“Signups are failing in production. Check the Vercel logs, reproduce it, and open a PR with a fix.”
Debug a Vercel App From Logs to Fix
See it run - On Mac
“Check my Supabase project's health: schema, row security and anything that looks off.”
Check Your Supabase Database Is Healthy
See it run
Do anything
Ask it a question or hand it a job, in any app.
Doing jobs for you is in the Mac app. Asking works on Mac, Windows and Linux.
