2 comments

  • Bender 1 day ago
    This seems like a complex solution to a simple problem. In sshd_config there are some directives that can close unused connections. [1][2] as of OpenSSH 9.2

    Another solution is to set a read-only shell timeout on servers and managed clients. This can annoy people causing them to do silly work-arounds so coordinate with your teams before applying this everywhere. Find an agreed upon and reasonable timeout that does not conflict with audit requirements that vary by company.

        readonly TMOUT=86400
    
    [1] - https://man.openbsd.org/sshd_config#ChannelTimeout

    [2] - https://man.openbsd.org/sshd_config#UnusedConnectionTimeout

    • mathgladiator 1 day ago
      For clarity, it is a mess of a solution, and the key features I wanted was to have a good overview of a variety of ssh sessions I was having open as I realized I was giving prompts to the wrong claude session hence I have have a name for every agent that isn't changing and also a sense if it is active or not. Basically, i wanted a dashboard for ssh.

      What I didn't realize that claude could build me my own little ecosystem. Every single thing here is vibe coded. The java terminal, however, kind of sucks. But, it solves my problem as my needs changed.

      If I was one of the cool kids, then I'd probably use Zellij and have that worker peon notification system set up... BUT, I'm happy to live in my own little bubble since I'm not (nor have I even been) a cool kid.

      • Bender 1 day ago
        I think if you are doing it as a learning exercise that is great. It is unfortunate that AI did not give the simple solutions.

        For what it's worth many of the "cool kids" I went to school with ended up failing at life so its probably not something one should aspire to. Your path of enlightenment is far more valuable. There's just something wrong with the thinking machines.

        • mathgladiator 16 hours ago
          There is something wrong with the thinking machines, I don't disagree... this is the new reality. I aim now to understand it.

          the AI gave me a solution that works for since I'm using Windows primarily these days (granted, the latest nonsense has me wanting to install linux, but every time I give in and go that route, I'm playing the meta game of managing linux. That was fine when I was a kid learning, but now I just want a machine that works. I evolved to just accept the defaults from windows, so I could move faster and adapted myself to that world.

          I was fairly content with my setup using IntelliJ and WSL2 except Microsoft crapped the bed and ruined performance for how disk access works between folders shared between Windows and Linux. This was the perfect set up for me, I got the OS that mostly works without too much headache or meta games plus a nice server-like environment that I'm comfortable with.

          Still, I had a situation that worked well enough and I could muddle on as the #1 reason I use IntelliJ is that that JVM is a magic box. I love the fact that I can run a unit test in IDE, and get code coverage like a junkie. Now, the cool kids will all go "eew, Java" or "eew, IDE" which I find hilarious because Java and the IDE let me retire doing infrastructure work, so I don't fit in to the zeitgeist with all the things that seem to break all the time.

          The key problem I was solving was more of matter of convenience where I wanted the titles of my terminals to stop changing since I was forgetting the context of each claude window. Combine that with claude using a tremendous amount of memory. I decided to rent a server to offload my work to a better machine with real linux, and then on a lark... I gave claude a single prompt that was like three paragraphs, and it popped out the first version of crank that worked the first time.

          It was eye opening. Now, was it great, no... but it worked and after a few prompts it shaped up into something usable. This, to me, is amazing. I have not read the code at all, and I have solved a pain point that I was having. Now, I do have new pain points, but for just working across 15 claude code instances all scoped to working directories for me, it's kind of amazing. I have a dashboard that gives me a new command view, and I'm blown away. I can annotate each terminal with some ideas of what I want to do. Does it have bugs? yes, but has it improved my velocity and made life easier. Yes!

          The game that I've spent my entire life playing has changed. Software can be extremely personal. Software can be tailored to me rather than learning another persons config or setup. I can have my own camp rather than buying in to some popular what ever or what not.

          My mind is blown on so many fronts, and I'm going to release everything I do. I'm going to vibe code an entire mountain of garbage with exactly one customer in mind: me.

  • atmanactive 1 day ago
    It seems like the Github ReadMe lacks the "Getting Started" section. I don't understand how to install/use this.