Inspiration
Installing AI Agents and their frameworks might become tedious very quickly. AgentLinux idea was basically coming from one question on my mind: "Could installing Codex and GSD framework be as easy as installing vim and glibc on Ubuntu?". I didn't find anything close to this, so I've built one.
What it does
AgentLinux is a Linux plugin, which helps you to provision different parts of your agent harness on your Linux distribution, like agent user configuration and installing agents themselves.
As a part of this hackathon, I've focused on one specific problem which I had as AgentLinux developer: automating QA.
Projects list of AgentLinux is rapidly growing and testing all projects becomes more complex and time consuming. Packages interact in very complex ways with each other, so testing effort grows exponentially.
So, therefore, instead I've decided automate QA Testing using Codex.
How I built it / Challenges I ran into
There is one big problem: it's hard to define the condition when the agent should stop.
While pondering on the problem, I've realized one thing. Testing process outcomes gradually decrease over time. In a first 30 minutes you can discovery 5 bugs. In next 30 - 2 more. In next - none. That's exactly what I've decided to try out as goal definition for my agent.
I made it iterate over fixed periods of time and only when it could no more find bugs in this new period of time, it would be allowed to stop testing. On top of that I've added one more condition that last 10 tests should surface no new bugs, so we could account for time-consuming tests too.
A result: a Codex skill which can be called at any desired point of time to conduct QA Testing for whatever scope that I specify.
Accomplishments that I'm proud of
Speaking of a results, using QA Testing workflow Codex surfaced 10 issues, five of which were genuinely new, it went through almost 130 test ideas in total and run in total run testing for around 11 hours.
Here are a couple examples worth showing. On the left, Codex managed to surface a conflict when both GSD and Codex are getting installed and Codex is getting broken because of incorrect hooks configuration. On the right, Codex managed to surface that Gemini CLI is getting retired so it has to be replaced with Anti-Gravity CLI.
What's next for AgentLinux
QA Testing workflow I've built is going to facilitate shipping all the further features on my roadmap for AgentLinux: automatic package updates testing and releasing, recommend AgentLinux package groups (so users would be able to install best toolings without need to keep track of all the advancements in the field, like rtk for reducing tokens), and extending further projects list included into AgentLinux.
Built With
- claude-code
- codex
- docker
- html
- qemu
- shell
- typescript


Log in or sign up for Devpost to join the conversation.