it runs every night on Linux, Mac and Windows, against whatever the agents released that day.
Except Cursor. The free plan doesn’t give you API keys, and I’m honestly not paying for a subscription just so a robot can test my plugin at 3 a.m. Cursor still gets checked by hand, by me. Old school
i make a plugin for coding agents. And for a while, every release looked like this: open Claude Code, type the same test prompt, check that the hooks fired, check the upload, check that the model actually got the rules
Then the same thing in Pi
Then Antigravity
Then Cursor 🧵
You're reviewing a PR with 5,000 lines of AI-generated code
All tests pass and everything works
The developer admits they haven't read a single line of it
Do you approve or reject?
And more importantly — why?
@oyibedev@blessedofcodes like ai?
English isn't my native language, so I have Grammarly installed to fix mistakes - it’s what adds the periods and corrects my texts
hey X 👋
looking to #connect with people who:
- live in Claude Code, Codex or Cursor all day
- build dev tools
- ship side projects with agents
- are building something solo from day 0
Say hi and let's connect 📷Building something?
what are you building? drop it below 👇
🚨 Perplexity just open-sourced SPACE, an AI agent sandbox.
They stress-tested 9 frontier models with 118 escape & breach attempts from isolated containers. 0 succeeded.
Meanwhile, devs are literally passing raw sudo access to autonomous terminal agents in prod without a second thought.
Are we building the next generation of software, or just speedrunning the biggest cybersecurity disaster in history? 🧵👇