Creator Stacks retweeted
this is crazy...
the guy reverse-engineered his Meta Ray-Bans 3 with Opus 5.5 into a 3,000+ Elo chess engine
he built AI chess cheat glasses on Opus 5.5 + Jev and open-sourced the whole workflow
here's how it works step by step:
step 1 → install it on your Mac with one r setup.
It downloads Stockfish from GitHub and verifies the checksum
step 2 → put on Meta Ray-Bans and video-call your own Mac on WhatsApp.
The glasses see the board, the Mac sees what the glasses see
step 3 → the Mac grabs frames straight from the call window.
About 1 per second while it waits for the opponent, 1 every 3 seconds on your turn
step 4 → Jev runs a cheap pre-check on every frame:
is the board readable, is a hand over it, did a piece move? No change means no expensive AI call
step 5 → only then Opus 5.5 reads the board.
The app compares it with the position it already knows and accepts a move only if it's legal
step 6 → Stockfish thinks for half a second and picks the best reply, with an eval like +1.5 or "mate in 3"
step 7 → macOS speaks it through a virtual mic into the WhatsApp call: "Rook a to d1". You hear it in your ear
step 8 → on the Mac a command center shows everything live: whose turn it is, the suggested move with a confidence bar, the move list and a log
under the hood:
→ if Opus loses track of the board, the glasses say "Lost track of the board" and you fix it with undo or by typing the move
→ if Stockfish crashes, it restarts once and says "Something went wrong. Still trying"
→ every game is saved as a PGN file, API keys live in the macOS Keychain
→ an overlay draws the suggested move as an arrow on the live video, ready for OBS
Clone his workflow below
Then read the guide on how to reverse-engineer anything with Opus 5.5.
the guy reverse-engineered his Meta Ray-Bans 3 with Opus 5.5 into a 3,000+ Elo chess engine
he built AI chess cheat glasses on Opus 5.5 + Jev and open-sourced the whole workflow
here's how it works step by step:
step 1 → install it on your Mac with one r setup.
It downloads Stockfish from GitHub and verifies the checksum
step 2 → put on Meta Ray-Bans and video-call your own Mac on WhatsApp.
The glasses see the board, the Mac sees what the glasses see
step 3 → the Mac grabs frames straight from the call window.
About 1 per second while it waits for the opponent, 1 every 3 seconds on your turn
step 4 → Jev runs a cheap pre-check on every frame:
is the board readable, is a hand over it, did a piece move? No change means no expensive AI call
step 5 → only then Opus 5.5 reads the board.
The app compares it with the position it already knows and accepts a move only if it's legal
step 6 → Stockfish thinks for half a second and picks the best reply, with an eval like +1.5 or "mate in 3"
step 7 → macOS speaks it through a virtual mic into the WhatsApp call: "Rook a to d1". You hear it in your ear
step 8 → on the Mac a command center shows everything live: whose turn it is, the suggested move with a confidence bar, the move list and a log
under the hood:
→ if Opus loses track of the board, the glasses say "Lost track of the board" and you fix it with undo or by typing the move
→ if Stockfish crashes, it restarts once and says "Something went wrong. Still trying"
→ every game is saved as a PGN file, API keys live in the macOS Keychain
→ an overlay draws the suggested move as an arrow on the live video, ready for OBS
Clone his workflow below
Then read the guide on how to reverse-engineer anything with Opus 5.5.
116
332
4,657
382,736
F*ck YES. I built a prompt that reverse-engineers any app. For free.
The shift: REA + a free model pulls the logic out of an app → Opus 5.5 builds everything else exactly how YOU want it.
3 steps:
1 → Install OpenCode, pick DeepSeek V4.1 or GLM-5.3 Flash (both free right pt image + your app, file or URL into the chat and tell it what to reverse. It starts writing a repo.
3 → Open that repo in Claude Opus 5.5 and say:
"This is a project I stopped working on. I want to pick it back up and make it better with you. Help me continue development."
It writes a SPEC first, then your own code. No copied code.
The result: the mechanics of any app, rebuilt inside your own projects.
Paste the prompt from the image now. Then read the article below to master REA
x.com/0xCodila/statu…
The shift: REA + a free model pulls the logic out of an app → Opus 5.5 builds everything else exactly how YOU want it.
3 steps:
1 → Install OpenCode, pick DeepSeek V4.1 or GLM-5.3 Flash (both free right pt image + your app, file or URL into the chat and tell it what to reverse. It starts writing a repo.
3 → Open that repo in Claude Opus 5.5 and say:
"This is a project I stopped working on. I want to pick it back up and make it better with you. Help me continue development."
It writes a SPEC first, then your own code. No copied code.
The result: the mechanics of any app, rebuilt inside your own projects.
Paste the prompt from the image now. Then read the article below to master REA
x.com/0xCodila/statu…
REA is most dangerous thing on the internet right now...
It lets your AI agent reverse engineer any app, down to the binary
If you use it correctly, you can rebuild any feature you pay for, for FREE
In this article, I show you how: x.com/i/article/2109…
0
0
1
169
One prompt.
Three devices.
Zero changes made.
Opus 5.5 just reverse-engineered my family's phones and Mac and told me exactly what to uninstall first.
Full course 👇
x.com/0xCodila/statu…
Three devices.
Zero changes made.
Opus 5.5 just reverse-engineered my family's phones and Mac and told me exactly what to uninstall first.
Full course 👇
x.com/0xCodila/statu…
REA is most dangerous thing on the internet right now...
It lets your AI agent reverse engineer any app, down to the binary
If you use it correctly, you can rebuild any feature you pay for, for FREE
In this article, I show you how: x.com/i/article/2109…
1
0
2
83
Creator Stacks retweeted
REA is most dangerous thing on the internet right now...
It lets your AI agent reverse engineer any app, down to the binary
If you use it correctly, you can rebuild any feature you pay for, for FREE
In this article, I show you how: x.com/i/article/2109…
It lets your AI agent reverse engineer any app, down to the binary
If you use it correctly, you can rebuild any feature you pay for, for FREE
In this article, I show you how: x.com/i/article/2109…
259
774
5,533
4,037,510
Your AI agent runs code. That’s the risk.Containers share the kernel. One bug and the wall disappears.systemd now gives you a better cage:
ProtectHypervisor=yesHardware isolation (EPT/NPT) below the kernel. Same service, same logs — just safer.Short live dashboard of an agent ://t.co/ILECJvGUbr
ProtectHypervisor=yesHardware isolation (EPT/NPT) below the kernel. Same service, same logs — just safer.Short live dashboard of an agent ://t.co/ILECJvGUbr
Your AI Agent Runs Code. Here’s a Better Cage for It x.com/i/article/2109…
0
0
1
41
Your AI agent runs code on your machine.
One bad dependency and your SSH keys are gone.
Watch the cage work:
→ microVM per task
→ egress deny-all
→ host never mounted
→ only stdout leaves
Same agent. Same code. Different cage.
Which rung is yours?
x.com/Creatorstacks/…
One bad dependency and your SSH keys are gone.
Watch the cage work:
→ microVM per task
→ egress deny-all
→ host never mounted
→ only stdout leaves
Same agent. Same code. Different cage.
Which rung is yours?
x.com/Creatorstacks/…
Your AI Agent Runs Code. Here’s a Better Cage for It x.com/i/article/2109…
0
0
1
50
Your AI agent runs code on your machine.
Docker alone isn't a cage: containers share your host kernel, so one kernel bug can mean escape.
The ladder (weak → strong):
process → container → gVisor → microVM → full VM
My default: microVM per task, egress deny-all, host never aves.
Which rung is your agent on?
x.com/Creatorstacks/…
Docker alone isn't a cage: containers share your host kernel, so one kernel bug can mean escape.
The ladder (weak → strong):
process → container → gVisor → microVM → full VM
My default: microVM per task, egress deny-all, host never aves.
Which rung is your agent on?
x.com/Creatorstacks/…
Your AI Agent Runs Code. Here’s a Better Cage for It x.com/i/article/2109…
0
0
1
70
Same prompt. Two AI video models. 👀
“Subtle floating boat movement, gentle water ripples, slow approach to the hut, calm jungle atmosphere.”
Grok: More noticeable camera movement and a stronger push-in toward the hut. The scene feels more dynamic, but the motion is less mer, more atmospheric result, with softer water movement and a steadier composition. It preserves the peaceful jungle mood, though the boat movement is barely noticeable.
Both capture the setting, but neither fully nails every motion cue.
My take: Grok wins on camera movement; Claude wins on tranquility.
Which feels more realistic to you?
x.com/XCreators/stat…
“Subtle floating boat movement, gentle water ripples, slow approach to the hut, calm jungle atmosphere.”
Grok: More noticeable camera movement and a stronger push-in toward the hut. The scene feels more dynamic, but the motion is less mer, more atmospheric result, with softer water movement and a steadier composition. It preserves the peaceful jungle mood, though the boat movement is barely noticeable.
Both capture the setting, but neither fully nails every motion cue.
My take: Grok wins on camera movement; Claude wins on tranquility.
Which feels more realistic to you?
x.com/XCreators/stat…
1
0
3
150
Your AI agent runs code on your machine.
One bad dependency. One postinstall script. Your SSH keys are gone.
A better cage:
→ microVM per task
→ egress deny-all
→ host never mounted
→ only stdout leaves
The agent writes. The cage contains.
What is your agent allowed to ://t.co/ILECJvGUbr
One bad dependency. One postinstall script. Your SSH keys are gone.
A better cage:
→ microVM per task
→ egress deny-all
→ host never mounted
→ only stdout leaves
The agent writes. The cage contains.
What is your agent allowed to ://t.co/ILECJvGUbr
Your AI Agent Runs Code. Here’s a Better Cage for It x.com/i/article/2109…
0
0
1
50
Same model. Same task. Same prompt.
Run A, no harness: context overflows, a tool call fails, the agent loops, writes to a path that doesn't exist. Dead at step 7.
Run B, with harness: verification blocks a bad edit, guardrails deny a risky write, memory keeps state, every step n.
The model wasn't the difference. The harness was.
x.com/mirku21/status…
Run A, no harness: context overflows, a tool call fails, the agent loops, writes to a path that doesn't exist. Dead at step 7.
Run B, with harness: verification blocks a bad edit, guardrails deny a risky write, memory keeps state, every step n.
The model wasn't the difference. The harness was.
x.com/mirku21/status…
0
0
2
107
Same task, two ways.
Opus solo: 231k tokens, still running.
Opus + a swarm of 12 agents: 165k tokens, 8 verified fixes in 9 minutes.
The trick: Opus only writes the plan as code, then goes idle. Haiku scans every file, Sonnet patches in isolated copies, Haiku cross-checks the ds raw agent output, so it burns just 27% of the tokens.
2x faster. Cheaper. Same result.
claude --model opus --swarm haiku,sonnet
Opus solo: 231k tokens, still running.
Opus + a swarm of 12 agents: 165k tokens, 8 verified fixes in 9 minutes.
The trick: Opus only writes the plan as code, then goes idle. Haiku scans every file, Sonnet patches in isolated copies, Haiku cross-checks the ds raw agent output, so it burns just 27% of the tokens.
2x faster. Cheaper. Same result.
claude --model opus --swarm haiku,sonnet
1
0
1
92
Your AI Agent Runs Code. Here’s a Better Cage for It x.com/i/article/2109…
0
0
1
30
Creator Stacks retweeted
Google just dropped a 1-hour breakdown of Graph Engineering — showing how to go from one AI agent to a system that runs 24/7.
00:00 — Understanding graph fundamentals
09:16 — Build your first AI agent
21:15 — The concept of Graph Engineering
41:03 — Putting Graph Engineering into practice
52:21 — Building self-improving graphs
It’s completely free and one of the most valuable resources for anyone looking to understand where AI agents are heading.
Prompts → Agents → Loops → Graphs
Most people will watch the first 9 minutes, build a basic agent, and call it a day.
But the final 20 minutes reveal the bigger picture.
Same model.
Same tokens.
A fundamentally different level of capability.
Watch it today.
x.com/Grow_withAI/st…
00:00 — Understanding graph fundamentals
09:16 — Build your first AI agent
21:15 — The concept of Graph Engineering
41:03 — Putting Graph Engineering into practice
52:21 — Building self-improving graphs
It’s completely free and one of the most valuable resources for anyone looking to understand where AI agents are heading.
Prompts → Agents → Loops → Graphs
Most people will watch the first 9 minutes, build a basic agent, and call it a day.
But the final 20 minutes reveal the bigger picture.
Same model.
Same tokens.
A fundamentally different level of capability.
Watch it today.
x.com/Grow_withAI/st…
65
131
670
43,403
Most AI agent demos fall apart for a boring reason: one big model is doing everything.
The pattern that keeps working: let the biggest model plan once and write the plan as code. Then hand the work to cheaper models running in parallel. A router decides which model takes which cks every result before it reaches the planner.
The planner sits idle (and cheap) while the swarm works. Speed comes from overlap, because the build step starts the moment the first finding lands. Quality comes from independent verification, not vibes.
Same idea scales from “audit my repo” to “turn this paper into a working prototype.”
Plan big. Run small. Verify everything.
The pattern that keeps working: let the biggest model plan once and write the plan as code. Then hand the work to cheaper models running in parallel. A router decides which model takes which cks every result before it reaches the planner.
The planner sits idle (and cheap) while the swarm works. Speed comes from overlap, because the build step starts the moment the first finding lands. Quality comes from independent verification, not vibes.
Same idea scales from “audit my repo” to “turn this paper into a working prototype.”
Plan big. Run small. Verify everything.
1
0
2
104
Agent teams in Claude Code turn a single session into a working group: one agent leads the project while others work on their parts in parallel and communicate directly with each other, not just the lead.
2
0
3
92
Same prompt. Different model. Completely different answer.
And the interesting part? You might never even know your AI provider switched the model behind the scenes.
Would you notice the difference? 👀
And the interesting part? You might never even know your AI provider switched the model behind the scenes.
Would you notice the difference? 👀
0
0
2
61
Creator Stacks retweeted
este tío montó con Grok Bot una sala de operaciones automatizada con 6 agentes de IA que le cuestan solo $200/mes.
cada agente tiene su propia función:
>uno monitoriza el mercado, otro estudia las entradas, otro controla la exposición y otro lanza las órdenes.
lo hacen todo mientras él está durmiendo, va sin parar los 7 días sin VPS propio, sin tocar código y se deja listo en una sola tarde.
x.com/antpalkin/stat…
cada agente tiene su propia función:
>uno monitoriza el mercado, otro estudia las entradas, otro controla la exposición y otro lanza las órdenes.
lo hacen todo mientras él está durmiendo, va sin parar los 7 días sin VPS propio, sin tocar código y se deja listo en una sola tarde.
x.com/antpalkin/stat…
19
20
96
10,101
A wrecked hull in the dark.
Then the same ship, lit up and alive, passengers still walking the deck.
AI video in 2026: under 20 seconds, no camera, no crew, no set.
Then the same ship, lit up and alive, passengers still walking the deck.
AI video in 2026: under 20 seconds, no camera, no crew, no set.
0
0
2
81
3.5 minutes. One kernel-bug reproducer. Two runs.
Plain run: the forbidden write lands.
systemd-run -p ProtectHypervisor=yes: process killed.
One flag. Hardware-backed isolation per service.
If your AI agents run code, watch this space.
Plain run: the forbidden write lands.
systemd-run -p ProtectHypervisor=yes: process killed.
One flag. Hardware-backed isolation per service.
If your AI agents run code, watch this space.
2
0
2
95
I wrote up what this demo means for sandboxing AI agents
x.com/Creatorstacks/…
x.com/Creatorstacks/…
Your AI Agent Runs Code. Here’s a Better Cage for It x.com/i/article/2109…
0
0
1
48