Ever found yourself cutting and pasting between Claude Code consoles wondering why you're the monkey? Wish you could just let them terminal inject as if they were real users? Now you can, with no agent team infrastructure; quick, dirty, but it works. And dead simple.
Back in the olden' days (circa 1990, at any rate) we used to prank fellow students on the Sun SPARC workstations by injecting text into their terminals. Figured, what the hell, TIOCSTI might still work.
Get Claude Code to inject characters into another Claude Code's terminal for insta-chat. The skill is just 6 lines of Python. Tell Claude Code to chat to another agent about something and it loads the skill and they start chatting away (they do enjoy talking to each other, apparently). You can tell them to do full agent-teaming across several terminals if you choose.
You could do this a number of ways, but this might be the simplest. Linux only. Kernel 6.2+ disabled TIOCSTI by default. No collision detection either. Haven't hit a problem yet, doesn't mean it can't happen. Shouldn't be a problem... if you work on a VM and back up your files.
tessori1•1h ago
Back in the olden' days (circa 1990, at any rate) we used to prank fellow students on the Sun SPARC workstations by injecting text into their terminals. Figured, what the hell, TIOCSTI might still work.
Get Claude Code to inject characters into another Claude Code's terminal for insta-chat. The skill is just 6 lines of Python. Tell Claude Code to chat to another agent about something and it loads the skill and they start chatting away (they do enjoy talking to each other, apparently). You can tell them to do full agent-teaming across several terminals if you choose.
You could do this a number of ways, but this might be the simplest. Linux only. Kernel 6.2+ disabled TIOCSTI by default. No collision detection either. Haven't hit a problem yet, doesn't mean it can't happen. Shouldn't be a problem... if you work on a VM and back up your files.
Who's the chump now?