You’re in control
Konduktor was built for engineers. It runs on your machine and is open source, so it’s made for poking around. The goal has been to make something that gets you up and running fast through a heavily-automated install script, a solid UI, and a fully-featured CLI, but there are aspects of configurability that haven’t been clearly exposed via those interfaces yet. The first thing you might consider tweaking is the task agent system prompt, and that can be done from the UI (in “Workspace Config”) or the CLI (konduktor config). We provide a simple yet solid default that works well but this is something advanced users will want to tune per workspace.
Beyond that, the next layer of configurability is the environment and the harness itself. There are plans to make things like skills configured from the Konduktor interfaces but for now the way to do it is pretty simple: SSH into your instance and manually install skills yourself. You can then tweak the task agent system prompt to mention these skills if you like.
Using the same method, you can also install anything else. This is the point of having Konduktor run on a remote machine, so you can more freely install things without worrying about polluting your local workstation.
Want Puppeteer to test in the browser? Want to setup a complex testing flow? Want to give the agent more tools to debug? You can simply SSH into the machine and install these tools.
There are a lot of plans for enabling easier configuration of environments for agents to operate in, but for now you can do anything, it’s just a bit more DIY.