All posts

Slack · Building in public

What it took to build ReactPing

The version that shipped is less than half the size of the one before it. Most of the work was deciding what to delete.

I ask Slack channels to react all the time: react if you're coming, react with your vote. Then the answers pile up on a message I've already scrolled past. So I built ReactPing, a Slack app that DMs you when someone reacts to your message, with who reacted, the emoji, a preview, and a link back.

It took four weeks and three versions. I built it with AI coding agents doing most of the typing, which made building cheap and made deciding what to build the real work.

Version one: it works on my laptop

The first version, on October 23, was a single Node service with SQLite, run on my laptop and exposed to Slack through a temporary tunnel. Within a day it delivered a real notification for a friend's reaction.

It also taught me how fragile that setup was. The tunnel's address changed when it restarted, Slack kept sending events to the old one, and a friend's reaction vanished without a trace. A prototype that only works while my laptop is awake is not something I can hand to anyone.

Version two: built for problems I didn't have

In November I rebuilt it for hosting on Vercel and Supabase, and I aimed high: email sign-in, a web dashboard, one inbox across every workspace you use, and a durable delivery queue with leases, retries, backoff, and a cron job to recover missed work.

It passed 85 tests and two rounds of code review. Then it hit a wall nobody had tested for: Supabase's free plan wouldn't let me change the sign-in email template without custom SMTP, so the sign-in flow couldn't work as designed.

That blocker was the best thing that happened to the project. It made me ask why ReactPing needed its own accounts at all. Slack already knows who you are.

Version three: delete most of it

On November 13 I cut ReactPing down to a Slack-only app. Your Slack sign-in is the whole account. Notifications stay inside the workspace where the reaction happened. There is no queue: Slack's request is acknowledged right away, the DM goes out right after, and if that fails, the notification is skipped rather than retried.

That one change removed about 5,800 lines and added about 2,700. Accounts, the dashboard, the cross-workspace inbox, the queue, the cron job, and the email problem all disappeared together. The tests went from 85 to 66, because the things they covered no longer existed.

From there it moved fast: live in my workspace on November 14, a second workspace on November 15, open source on November 16, and a message preview, a feedback form, a demo video, and its own domain by November 17.

What went well

Making privacy checkable instead of promised. ReactPing asks for two permissions and stores no message text. Its security page links each claim to the line of code that enforces it, and a test fails if that code moves. If I say it, you can check it.

Testing with real people early. Friends reacting in a real workspace caught things a mocked test never would, and every claim of "it works" was checked against a real DM.

Removing features without regret. The day I built a way to send feedback by DMing ReactPing, I removed it, because it needed an extra Slack permission. A website form does the same job without asking for anything.

What didn't

I designed for scale before I had a single user. The queue, the retries, and the cross-workspace inbox cost weeks and were deleted before anyone used them.

A testing switch took the whole site down. ReactPing has a setting that lets me get notified about my own reactions while testing. On November 17, I entered a value in the dashboard that the code didn't accept, and the app treated it like missing configuration: every page and Slack event returned an error until it was fixed. Now anything unclear means off, and only the settings ReactPing can't run without are allowed to stop it.

I also skipped testing Disconnect and uninstall on the first release because I didn't want to break the working install. I tested them two days later and they worked, but "skipped" sat in the docs longer than it should have.

And the free tier has a known risk I chose to accept: Supabase pauses idle free projects, and if that happens, notifications stop until I restore it. I'd rather be honest about that than hide it.

What I'd do differently

Start with the smallest version that can reach a real user, and add infrastructure only when a real problem asks for it. Version three is what I should have built in October.

ReactPing is free forever and open source. If you live in Slack, try it and tell me what breaks.