Agent-Board
A live pixel-art view for Claude Code agents. Every subagent is a small figure that walks into the room of the tool it is using. At a glance you see who is working and who is waiting for approval.
AI agent · public, release v1.4.0
Stefan Paun
I build AI agents and I’m learning cybersecurity.
Agent-Board shows live which of my Claude Code agents is working and which one is waiting for my approval. Open source, built by me.
17 · high school student in Vienna · graduating 2027, then computer science
Go to projectEnter opens · Escape closes
Skip to projects
Every coloured line is a project. Tap one.
Yellow is AI agents, cyan is security. Red is learning games.
3,619 tests. Automated tests for Jarvis, every night in the cloud.
What I’ve built.
Every project with a real number.
Jarvis
My voice assistant for Mac and Windows: press a key or say “Hey Jarvis”, it gets things done with its tools and answers by voice. Data encrypted, key in the macOS keychain, a test run in the cloud every night.
AI agent · private, used every day No public code, because personal data is involved.Security
I checked my iPhone against 6,630 public spyware indicators with Amnesty International’s Mobile Verification Toolkit: none found, but that is not proof. Plus 6 rooms of the Pre Security path on TryHackMe and Cisco’s Introduction to Cybersecurity course.
Security · in progress I don’t have a certificate yet.Rally engine
A pixel-style learning game with a world map, bosses, a shop and a mock exam. Built once, turned into two games: the GWB Rally (geography and economics) with 273 questions and the Cyber Rally with 135 questions and Leitner-style spaced repetition.
Learning game · 408 questions in two games · not public yet
Claude Code tools
A guard checks every commit and push against 10 patterns before it happens, plus a cockpit for open pull requests and skills for recurring workflows.
AI agent · 2 mods, 2 skills The guard is open source, the cockpit and skills are private.How I build. Every project goes through the same layers. Whatever fails a layer doesn’t move on.
Plan
I write down what gets built and split it into phases.
Agents
Claude Code agents build separate parts in parallel, cloud routines keep working at night. I review and merge.
Tests
Every change runs through the tests before it stays.
Security
No keys in the code, no private data. Before publishing I check which attack paths exist.
Release
Only finished things go public: with docs, a license and a demo.
Now. As of .
- Agent-Board now has a city view: every agent walks through a small pixel town. New since .
- Security: the Pre Security path on TryHackMe and the Cisco course are in progress.
- School: preparing for my final exams (Matura), partly with my own learning games.
Notes (German)
- How I build with Claude Code agentsworkflow and what went wrong
- What tests taught melessons from Jarvis and Agent-Board
What I stick to.
- Proof over adjectives. Every project shows what it does and one real number.
- Honest about my level. I’m 17 and learning in public. Not a senior, but I ship.
- Privacy first. Even this page reveals nothing personal that doesn’t belong here.
- Fast and light. A page without a framework that also runs in LinkedIn’s browser on a phone.
Contact.
Questions about my projects, feedback or working together: GitHub is the fastest way.
Issues and feedback on Agent-Board are welcome on GitHub.