I containerize Claude Code in k8s.

It is tasked to monitor a namespace and in the unfortunate case of application errors, it is tasked to do a hotfix and document it. It succeeds.

Basically a 24/7 on-call engineer.

Repo, examples & results below:

denislavgavrilov.com/p/clopus-watch… Here is the architecture. It consists of a dashboard & a cronjob with config env vars for auth_mode (api_key/auth.config) and worker_mode (autonomous/watcher). It only monitors namespaces you particularly set for it to monitor.

I successfully ran a Claude Code instance for 24-hours without any action from me.

I gave it long(qdrant) and short(sqlite) -term memory, and access to a browser.

The run generated:
- 500 projects
- ~450k LOC
- 20 long-term records
- 1500+ screenshots

denislavgavrilov.com/p/clopus-02-a-… This is a quickly drawn architecture of how the system works.

Basically, there are two ClaudeCodes living in TMUX panes. One is the builder, the other one is the "watcher". There is a systemd timer that sends the string "CHECK" to the watcher, and it checks on builder claude /w tmux send-keys commands.

I gave Claude full access to the / directory of a Linux virtual machine.

I then asked it to create a "child" Claude that it can talk to and do anything together.

I gave it a domain and it figured everything out on its own. You can watch the stream here: clopus.live And the "brain" approves commands the "child" asks for :) this is amazing

Hopefully it does not stop, and stays in an infinite loop until I lock my account out due to limits

BOTTING: How I managed to get access to the private X API (API V2) entirely for free & a short dive in Web Development

A thread (0/12)
00/12: Pre-Word

This thread is about
- how bots work
- session hijacking
- how I used the X API V2 for free
- examples (browser extension and a terminal client for X)