Avi & Shay's Stealth Startup

re_gent crossed 2,000 clones, 688 stars, and 6 contributors. Nice numbers. Finally people are starting to say this out loud: @github is not enough for AI-written code. It can tell you what changed. It cannot tell you what prompt caused the change, what tool call touched the file, which agent session produced it, or which now-deleted conversation explains the diff. We gave agents write access to codebases. Then we let the context disappear. re_gent is our first public answer to that problem. Devs are already using it daily. The feedback has been kind of wild. And the best part: This is only the small public piece of what we’ve been building underneath. Big thanks to @AnandChowdhary for the shoutout. #ai #versioncontrol #re_gent

"Look how badly it writes this. I would have done it so much better." That’s the sound of our CTO yelling at his monitor again. We asked an AI agent to fix a tiny bug. It fixed the bug. And then it deleted a chunk of code it wasn't even asked to look at. (Because why not? It's free real estate). This is the reality of AI coding agents right now. They are fast, confident, and have absolutely zero boundaries. Right now, the industry's solution to keep them in check is hilarious. We just dump a giant architecture.md file into the prompt and pray it understands the system. But dumping static text files to explain a living, breathing deeptech architecture is like giving a toddler a chainsaw and a map drawn in crayon. Until the industry figures out how agents can actually understand context, we’re all just babysitting very fast, very arrogant junior dev.

The challenge is live. After a lot of requests from participants, we’re happy to announce: Private Challenge Circles are here 🥳 You can now create a private challenge, invite friends, and compete together in your own room. Your score still counts in the global pool, so you can beat your friends and still fight for the leaderboard. This one is tough. This one is nostalgic. Go win. theincidentchallenge.com

The challenge is live. After a lot of requests from participants, we’re happy to announce: Private Challenge Circles are here 🥳 You can now create a private challenge, invite friends, and compete together in your own room. Your score still counts in the global pool, so you can beat your friends and still fight for the leaderboard. This one is tough. This one is nostalgic. Go win. theincidentchallenge.com

The challenge is live. After a lot of requests from participants, we’re happy to announce: Private Challenge Circles are here 🥳 You can now create a private challenge, invite friends, and compete together in your own room. Your score still counts in the global pool, so you can beat your friends and still fight for the leaderboard. This one is tough. This one is nostalgic. Go win. theincidentchallenge.com

The challenge is live. After a lot of requests from participants, we’re happy to announce: Private Challenge Circles are here 🥳 You can now create a private challenge, invite friends, and compete together in your own room. Your score still counts in the global pool, so you can beat your friends and still fight for the leaderboard. This one is tough. This one is nostalgic. Go win. theincidentchallenge.com

The challenge is live. After a lot of requests from participants, we’re happy to announce: Private Challenge Circles are here 🥳 You can now create a private challenge, invite friends, and compete together in your own room. Your score still counts in the global pool, so you can beat your friends and still fight for the leaderboard. This one is tough. This one is nostalgic. Go win. theincidentchallenge.com

The challenge is live. After a lot of requests from participants, we’re happy to announce: Private Challenge Circles are here 🥳 You can now create a private challenge, invite friends, and compete together in your own room. Your score still counts in the global pool, so you can beat your friends and still fight for the leaderboard. This one is tough. This one is nostalgic. Go win. theincidentchallenge.com

The challenge is live. After a lot of requests from participants, we’re happy to announce: Private Challenge Circles are here 🥳 You can now create a private challenge, invite friends, and compete together in your own room. Your score still counts in the global pool, so you can beat your friends and still fight for the leaderboard. This one is tough. This one is nostalgic. Go win. theincidentchallenge.com

The challenge is live. After a lot of requests from participants, we’re happy to announce: Private Challenge Circles are here 🥳 You can now create a private challenge, invite friends, and compete together in your own room. Your score still counts in the global pool, so you can beat your friends and still fight for the leaderboard. This one is tough. This one is nostalgic. Go win. theincidentchallenge.com

we built a Claude Code plugin implementing harness self-evolution from "Harness Updating Is Not Harness Benefit" (Lin et al., 2026) *the bottleneck is following skills, not writing them install the plugin from: github.com/Stealthy-McSte… based on arxiv.org/abs/2605.30621 written by @dakuowang @cihangxie @yuyinzhou_cs

We added a leaderboard to the incident challenge 💥 Does this mean production debugging is now officially a competitive sport?

And the winner is... 🥁 This challenge was not easy. Hundreds of engineers participated, and the results were seriously impressive. The difference between #1 and #2 was 13 seconds. Here are the top 5: #1 Kyle H. 👑 #2 Paul Romeo #3 Jamie Wood #4 Thammaiah K M #5 Davi M. Huge respect to everyone who took part. Next challenge drops in two weeks. Bring someone you want to beat 😊

The Challenge is Live - and is on Product Hunt! Let's gooo. 1. Support us on PH 🫡: producthunt.com/products/the-i… 2. Start the challenge (this one is especially tough) : theincidentchallenge.com

@copyconstruct That's exactly why we've launched The Incident Challenge - it's a debugging game for engineers using AI. Devs that only rely on AI, will not win.

If you believe “humans should deeply understand what they’re building with AI”, it follows that AI generated code should be optimized for “human cognition” Yet, often AI written code is actively hostile to this goal, and worse, a ton of “best practices” encourage this hostility.

@AnandChowdhary Thanks for the shoutout 😍

Agent version control feels obvious once you see it. Git tells you who changed a line. @stealthyishere's re_gent tries to answer the missing question: which prompt, tool call, and agent session wrote this line? That is the audit trail I want before agents touch serious codebases. The funny bit is that /compact made this problem visible. We gave agents write access, then lost the conversation that explains the diff. A content-addressed log for agent steps feels boring in exactly the right way. github.com/regent-vcs/re_…

re_gent is live on Product Hunt, currently #5. Help us push it to #1! Why re_gent was built: Git tells you what changed, but not.. - What the agent did to get there. - Not which prompt caused it. - Not what the agent did along the way. - And definitely not after /compact. What re_gent does: • Traces what your agent did, step by step • Blames a line of code back to the prompt/session that caused it • Rewinds agent work across code and conversation • Keeps history even when sessions get compacted or fragmented Apache 2.0, free forever, open source. We’d love your support 👇 producthunt.com/products/re_ge…

🚀 Introducing re_gent: version control for AI agent activity. Git shows you what changed in the code. re_gent shows you what your agent did to get there. 👾 Agent blame (which prompt/session caused a line or change) ❮❮ Rewind (roll back agent work, even after compact) 🔍 Trace (follow what your agent did, step by step). re_gent is open source, Apache 2.0, and free forever. Purely for the community. Check it out! Repo: github.com/regent-vcs/re_… Website: re-gent.dev

🚀 Introducing re_gent: version control for AI agent activity. Git shows you what changed in the code. re_gent shows you what your agent did to get there. 👾 Agent blame (which prompt/session caused a line or change) ❮❮ Rewind (roll back agent work, even after compact) 🔍 Trace (follow what your agent did, step by step). re_gent is open source, Apache 2.0, and free forever. Purely for the community. Check it out! Repo: github.com/regent-vcs/re_… Website: re-gent.dev