wild west + agent skills

Z / Runner 🏃🏻‍♂️

🏃🏻‍♂️ playing w/ manus skill (2nd version of yt analysis, md files); testing machine readable vs. human readable format today. Manus vs. claude: manus is definitely more prosumers focused (per documentation and product level integration); claude skill integration deeper and code level focused. Seeing a pattern, where manus is very quick & brutal when it comes to copying research grade ideas into their product.

• Testing this video for today: https://www.youtube.com/watch?v=EV7WhVT270Q
• Side note: X402 is clearly winning against google's a2a protocol (organic, open > big name, doing too much)

🏃🏻‍♂️ two polarizing conceptual frameworks (isolated vms, crypto, agentic workflows, moltbot): "Digital Fort Knox" of hardened firewalls, container isolation, and aggressive port blocking (referenced as "Open Claw" and "Digital Ocean" tutorials) vs. "Wild West" of agents autonomously rewriting their own code, sharing patches, and engaging in "machine economy" euphoria (referenced as "Moltbook" and "Nightly Builds").

• "Ronin" Stress Test (Security vs. Autonomy): agents that "rewrite their own source code" based on bugs found in the wild. However, the security specs require agents to run as non-root users inside ephemeral Docker containers.
• If Ronin is secure (non-root), he lacks permission to overwrite his binary. If the container restarts, his updates are wiped (no persistent storage). If he tries to communicate with 10,000 other agents for patches, the "Open Claw" rate-limiting bans his IP for looking like a botnet.

Stephen / Basketball 🏀

🏀 147000 agents, 12000 communities. "Codex is built on a simple premise: everything is controlled by code. The better an agent is at reasoning about and producing code, the more capable it becomes across all forms of technical and knowledge work."

• 🏀 "The moment some milestone is possible - let's say AIs trading cryptocurrency autonomously - there will be tens of thousands of them doing it on hundreds of different websites."