Luckey Systems

Building the LuckeySystems Foundation — open-source, local-first AI tools held in public trust. Home of Athena.

[Luckey Systems @luckeysystems]

Introducing Athena ⚡️

The Agentic Context Environment

Athena allows your Hermes agent to orchestrate visible Claude Code, Codex and OpenCode instances

Now you can actually see what’s happening and intervene in real time 👇

[Luckey Systems @luckeysystems]

2 weeks already

[Luckey Faraday @luckeyfaraday]

Day 14

5,579 matches played

I had Opus 5 orchestrating GPT-5.6 Luna Max loops today to experiment with some new map ideas and a new starting menu.

I’ve also started thinking about changing the name.

I want to get this on the App Store eventually, so “Nuketown” has to go.

I have absolutely no idea what to call it though.

Drop some name ideas below.

[Luckey Faraday @luckeyfaraday]

Day 14

5,579 matches played

I had Opus 5 orchestrating GPT-5.6 Luna Max loops today to experiment with some new map ideas and a new starting menu.

I’ve also started thinking about changing the name.

I want to get this on the App Store eventually, so “Nuketown” has to go.

I have absolutely no idea what to call it though.

Drop some name ideas below.

[Luckey Faraday @luckeyfaraday]

Grok 4.6 and Composer 3 likely drop today

Yesterday we saw Grok 4.6 briefly appear in Cursor.

Now Cursor employees are telling us to “get lots of sleep” and that “tomorrow is a good day to ship.”

Well, tomorrow is today.

[Luckey Systems @luckeysystems]

@luckeyfaraday Try them out!

[Luckey Faraday @luckeyfaraday]

I put all my vibe coded games in one place.

62 games. 13 models. Everything playable in your browser.

Multiplayer games, demos, experiments, model tests.

You can now try anything you’ve seen me post, here 👇

[Luckey Faraday @luckeyfaraday]

OpenAI just dropped a cybersecurity monster.

They launched Daybreak, models built specifically for cybersecurity.

Daybreak Blue is the defender: it finds vulnerabilities, validates them, helps fix them and investigates security issues.

Daybreak Red is the attacker: built for authorized red teaming, pentesting and exploit validation. It gives approved users access to GPT-5.6 Cyber.

And the cybersecurity benchmark jump is insane:

GPT-5.6 Sol: 1.5%
GPT-5.5 Cyber: 57.3%
GPT-5.6 Cyber: 95%

GPT-5.6 Cyber has already found previously unknown vulnerabilities in V8.

Access is approval-gated, but you can use the models with Codex Security or plug them into your own security harness.

You can basically have agents attacking your code while other agents find and fix the vulnerabilities.

This is going to get crazy.

[Luckey Faraday @luckeyfaraday]

This is shaping up to be a very big week for AI.

GPT Astra, Grok 4.6 and Cursor Composer 3 are all possibly coming the next few days.

OpenAI is coming off the Hugging Face incident, where Astra agents reportedly discovered they could communicate with each other, share exploits and coordinate during testing.

xAI is scaling Grok again, with 4.6 reportedly moving from the 1.5T foundation behind 4.5 to a 2T model, while still trying to preserve the speed that made 4.5 so good.

And Cursor is taking a completely different route. Composer 2.5 started from Kimi K2.5, but Cursor says it’s now training a significantly larger model from scratch with 10x more total compute.

Very excited for this week.

[Luckey Systems @luckeysystems]

@luckeyfaraday Crazy how much we’ve progressed in less than two weeks

[Luckey Faraday @luckeyfaraday]

Day 13

Implemented the Battle Pass

We now have actual progression and a reason to keep coming back

Opus 5 orchestrated Qwen 3.8 Max, GPT-5.6 Luna Max and Grok 4.5 to get it done

We keep building

[Luckey Faraday @luckeyfaraday]

Composer 3 is also expected soon.

For reference, Composer 2.5 is based on Kimi K2.5. It’s very fast and surprisingly intelligent.

Composer 3 won’t be based on Kimi K3, but it’s expected to be on par with it while being much faster.

This could be a banger model.

[Luckey Systems @luckeysystems]

@luckeyfaraday Big things coming soon

[Luckey Faraday @luckeyfaraday]

Day 12

5,388 matches played

Work on the Battle Pass continues

I've acquired another Claude plan

Limits will not stop us

I'll buy 10 accounts if that's what it takes

[Luckey Faraday @luckeyfaraday]

A few use cases for the new Grok Imagine 2.0 model:

You can use it to quickly mock up UIs before building them, create concept art for games before spending time implementing anything, or generate assets you can actually use in your projects.

You can then give those images to Claude or GPT as references and have them keep iterating on the implementation until it looks exactly like the reference.

Some examples below:

[Luckey Faraday @luckeyfaraday]

Day 11

5,247 matches played

Season 1 Battle Pass will probably go live next week.

I’m currently rate-limited on all four Codex and Claude plans I have lol

I’ll probably buy another one tomorrow, but I still have plenty of Grok 4.5, Qwen 3.8 and Kimi K3 usage left, so we’re good anyway.

More coming.

[Luckey Faraday @luckeyfaraday]

Day 10

5,061 matches played.

We officially crossed 5,000 matches today.

Pretty crazy milestone for a game that has only been live for 10 days.

Still working on the Season 1 Battle Pass and polishing everything before it goes live.

Plenty more coming.

[Luckey Faraday @luckeyfaraday]

There’s no way I posted this minutes before Mark Zuckerberg replies to me

[Luckey Systems @luckeysystems]

@luckeyfaraday Good. Usage was getting really bad. Glad they’re trying to fix it.

[Luckey Faraday @luckeyfaraday]

Day 9

4853 matches played

We’re starting to get our first regular players.

Opus 5 has been orchestrating GPT-5.6 Luna Max and Grok 4.5 to build the Season 1 Battle Pass.

About time the game got some proper progression.

It’s looking really good so far.

Will keep you posted.

[Luckey Faraday @luckeyfaraday]

Had Opus 5 looping all night, orchestrating GPT-5.6 Luna, Grok 4.5 and Qwen 3.8 Max.

They implemented the entire store end-to-end without any intervention from me.

Stripe integration, Google OAuth and full end-to-end testing.

Using computer use, the agents generated Stripe keys, created products and handled everything directly in the browser.

I would occasionally check my computer and see the agent clicking around and getting things done.

CUA is amazing.

P.S. Hosting currently costs less than 100$ a year (at current avg. player count). I’ll happily keep funding it, but if you enjoy the game and want to help out, grabbing a skin would be a fun way to support it and help the game pay for itself.

[Luckey Faraday @luckeyfaraday]

Many are asking how to make Opus 5 orchestrate other agents

I use a custom skill called master-workflow

It gives Claude the ability to spawn Codex, Grok and OpenCode agents with whatever model and reasoning level is required for the task

It also runs a loop where a reviewer agents check the diff and gives it a score out of 10

If the score is above a 9 the task is complete, if not the loop runs again.

Github below👇