Agents don’t just help build features. They can help try to break them, too.
In our latest GitHub Podcast episode, Sameen Karim takes us behind the scenes on building stacked PRs. Listen to it wherever you get your podcasts or at youtu.be/sbHoDdAalMI?si…
In our latest GitHub Podcast episode, Sameen Karim takes us behind the scenes on building stacked PRs. Listen to it wherever you get your podcasts or at youtu.be/sbHoDdAalMI?si…
Oct 19, 2026 · 7:36 PM UTC
32
13
105
29,307
@github You promote agents, then suspend a PAID Pro user for using them. ZERO warning, ZERO reason. ~400 research repos locked. Restore my account. Now. x.com/wan73184085/st…
PSA if you use AI coding agents on GitHub (Codex, Claude, Cursor, Devin):
GitHub just suspended my account. No warning. No email. No reason.
400 repos are gone.
Harvard researcher, paid Pro plan, 100% legit work.
You could be next. 🧵 @github @GitHubCommunity
0
0
0
39
@github having an agent try to break my PRs sounds pretty great
0
0
0
20
@github only if it's a different agent. the one that built the feature is the worst one to try breaking it.
0
0
0
59
@github Using agents to try to break a feature before it ships is an underrated use. Did the stacked PRs team run those adversarial agents in CI, or as a separate review step?
0
0
0
20
@github Using an agent to try to break each stacked PR before human review sounds useful. The harder part seems to be keeping the test signal stable as lower layers change. How do you handle that in practice?
0
0
0
11
@github give the second agent the requirement and the diff, then ask for a failing test. “looks good” is a very cheap answer.
0
0
0
6
@github Stacked PRs with agents sounds great until review lands on the bottom PR. Does the agent restack everything above it, or is that still on me?
0
0
0
42
@github Stacked PRs and agents are a natural fit. An agent can split a big change into small layers, and a second one can attack each layer with tests before a human reviews it. The painful part is still the rebase cascade when feedback lands on the bottom PR.
0
0
1
62
@github Stacked PRs give the agent a useful place to attack: the seams between changes. Test each PR against its declared parent, then the whole stack together. A green top-of-stack build can hide a middle PR that won't stand on its own when merged or reverted.
0
0
0
64
@github Agents trying to break things is just chaos engineering that never sleeps. Tempted to let one loose on my homelab.
0
0
0
47
@githubthat try to break things are the ones I end up trusting. I already make my agent review its own pull requests before I look, and it catches its own slop more often than I expect. Stacked PRs are the perfect stress test though. If an agent can build on its own half finished work without losing the plot, that is real progress.
0
0
0
31
@github agents testing code before review is a very polite plot twist
0
0
1
38