put my tiny eval harness on GitHub. 20 messy inputs, expected outputs, and one command to check whether a prompt change actually helped. boring tools win.
A Chrome extension, on your side of X.
My X. My rules.
Hide spam replies. Keep the conversations you came for.
Find your next good read, with a little help from Ownfeed.
the thread.↘


here. AGAIN?↙
A cleaner feed.
Your kind of X.
The curious builder.
“More real AI projects. Less politics in my feed.”

Sample demo
A useful start.
A better fit over time.
You don’t need to explain yourself all at once.
Tell it what you want. Refine it as you go.
- 01Say what you want less of
“Less promotion. Keep the technical discussion.” Start with something specific.
- 02Make room for better finds
Ownfeed looks for useful reading around your interests and what matters to you now.
- 03Keep the last word
See what it remembers. Correct a decision, show a post once, or undo a change.
What you watch
isn’t always
what you want to see.
A long scroll isn’t always a good one. What you ask for, choose to hide, and want to learn matters more.

Make X worth
opening again.
Cleaner replies. Topics you choose. People you keep.
Ask for early accessFor Chrome on desktop. Public installation is being prepared.
Good threads deserve
better replies.
one useful change to my agent: it has to run the tests before it can say “done”. better than any prompt tweak this weekend.
is the test runner part of the loop, or a separate check at the end?
inside the loop. if a test fails, the agent gets the failure output and has to try again.
same here. saving the failing cases as regression tests helped more than changing models.