Connect GitHub on the cursor dashboard or in GrokBot plugins. Cursor is best.
Create a private repo or have cursor do it. Then have cursor install what it needs for cloud agents.
Tell GrokBot what you want to do and to have cursor cloud agents do it. It will dispatch cloud agents. If you have cursor desktop, GrokBot will give you a link to pull it up in cursor and you can watch that bot.
I’ve seen the shit they do. If it takes so many rounds to figure out what word or phrase it’s hanging on to incorrectly, how many returns would you need to make before it got the right socks?
Screw popularity, you’ve just got common sense.
Trust is earned, systems must prove 1st
Oh. Good one!
A comedian with no sense of humor. That’s rich. 🙄
Look, I don’t disagree with the point, I disagree with the delivery. You make far too many assumptions in one blanket inflammatory give me attention post. Well, you got it. Good job.
Ai is a tool. So is photoshop. Not all who use photoshop are artists or capable of quality work, but many artists can do amazing work in it. And I mean old school pre-CS generations here especially. Don’t want to confuse it with new Ai features.
I disagree in the framing as your statement is like saying all photoshop is slop and anyone who uses it is…
It’s a tool. The quality of output depends on the user and the goals.
If you want to rail on slop, be specific. Otherwise, you come off a little like tilting at windmills here.
Have a nice day Don Quixote.
@twid For sure. It’s been a blessing. I finally prioritized some Hermes projects and have been putting Grok through its paces. Tailscale on GrokBot is so much better than Codex remote. Hands down, no contest.
It is possible. Espressif has official USB ECM host drivers for a few specific Ethernet dongles that allow it to use Ethernet. By default it just recognizes the USB and can be a USB host, but lacks a driver for Ethernet use. So unless you just enjoy writing driver stacks or porting, check the supported chipsets first and you’re golden.
The human is the one in the driver seat. If the human doesn’t understand or know how other systems work, the most efficiently leveraged will be what is familiar. That makes sense for work that needs done. What you suggest is a prime opportunity to learn and experiment.
I would guess that most, not all, are obsessed with trying to monetize or get ROI and not spend just for new knowledge. 🤷🏼♂️ just a guess.
There’s also advantage to aligning with training. Unless the bots are trained on other systems, the structure will require much more explanation and more rules. Using the same CoS model allows for slop setup and simple role explanations to work very well.
Hard stop is damaging to us. Keep running is how some will abuse the gift for free work.
I’d propose a soft stop. @1% agent is alerted and at 0% it’s instructed to not stop mid line, but to finish the current task, note the stopping point and then end gracefully. That’s common ground with mutual respect IMO.
Been working with Grok @bot a bit today. My wife looks over and just says “it looks like a toilet bowl”
I was confused at first. Now I can’t unsee it, and neither can you. You’re welcome.
@johnennis To be clear, I mean saved data on disk. The file store is protected through resets. I don’t mean data on any billable or accounting scale. Data = GB saved files.
There are computer reset options in settings. It does not affect data or cloud configured plugins through cursor, but will wipe any installed applications or MCP’s or whatever you’ve built. That will require rebuild from files or downloads.
They occasionally reset the computers. I have found it handy to build a rehydration script it stores in the data. After a reset, I have it run the rehydration script which can rebuild the working image I want on the machine.
You can also build custom scripts for different tasks. If you want to switch to different work, you can manually reset the computer and rehydrate from a different script to load up a different set of tools.