A VR Walk-through of https://t.co/EBoWtIbI4q
I created a wire-frame live version of the arcade for my lower powered phone so I could walk around capturing accelerometer data & my voice, uploaded to my PC, then replayed in the live #arcade on my fast PC & recorded.
#PACMAN
Pro AI Tip:
When you deploy a library, INCLUDE docs directed to AI agents. Have your AI agents write those docs. Have it put them in best practices file names and locations.
This makes your tools EXCEPTIONALLY more valuable to your users that are using AI.
Include gotchas.
@pixel2ai I first switched to T-Mobile for many years. Now on Patriot Mobile. It uses AT&T, T-Mobile, and Verizon towers for the best signal. Costs less, and doesn't re-subscribe me to services I've explicitly told them to block (that's what Verizon did to me EVERY month).
Pro AI Tip:
Claude Code ends long replies with a TL;DR because it's in my rules file. When it forgets, a Stop hook catches the reply and sends it back. It costs nothing when it's done right and one extra turn when it isn't, which beats re-reading every long reply myself.
Pro AI Tip:
Reading a PDF raw costs Claude Code about 1.2-1.6k image tokens per page. A 10-page PDF measured ~12,000 tokens raw vs under 800 as markdown. A hook now converts PDFs and Office docs to markdown, caches them, and points Claude Code at the cheap version.
Pro AI Tip:
Claude Code can't read a whole file over 300 lines anymore. A hook denies it and offers two options: read just the slice it needs, or hand the path to my local LLM, which reads the file itself so the contents never enter Claude's context.
Pro AI Tip:
Claude Code once sat all night waiting on a permission prompt for a single raw ssh command. Now a hook lets it run only scripts and git. Anything else is denied with the reason attached, so it fixes the command itself instead of waiting on me.
Pro AI tip:
A Claude Code session once deleted someone's entire C: drive when a quoting bug collapsed a path to the root. My delete-guard hook reads every script a command runs, checks every delete target, and blocks anything outside the project. No override, and it fails closed
Pro AI Tip:
Claude Code once wiped a whole ticket's work with git clean, and once killed an unrelated CI build with pkill -f. Now a hook blocks those, along with git reset --hard, kill -9, unbounded rm -rf and whole-tree reverts. Each block explains the safe way to do it instead.
Pro AI tip:
My AI coding agent once killed itself: it restarted the local proxy its own API calls ran through and died mid-task. Now a hook blocks any command touching that lifeline (proxy, base URL, API keys, auth). It can look, never change.
Pro AI tip:
My AI coding assistant never sees my secrets. A hook hard-blocks SSH keys, .env and credential files from its context. Anything that needs them runs locally, in scripts I run myself or on a local LLM. Even my home IP is written server-side and never printed.
@ExactlyJosh Years, unless you count the web version built into MS devops & Teams. I've been using LibreOffice for local docx files and now I'm using WordCraft.
@SuiBuilds Beginners do. Pros have multiple agents running for multiple projects. Some days I spend the whole day alt+tabbing between them to respond to the ones waiting on me.