I rebuilt my site today with the focus on folders over agents. It's geared toward providing the workflows for businesses. Not sure if this is allowed here, but I'm sharing anyway. Take it down if it breaks any rules. So many great minds here, so I thought it was worth risking criticism or advice on how to improve. Perhaps it will give you ideas for your own site. Check it out: optimarketai.com

I have directed film sets for years. Last night I directed a website the same way, and I never wrote a prompt. On a set you do not operate the camera. You watch the frame and call what it needs. Slower. Dirtier. Again. The crew has the hands. You have the eye. I ran a build the exact same way. A live page open on screen, an agent holding the hands, and me reacting out loud to what moved in front of me. One change. Watch it land. Call the next. DIRECT THE SCREEN, DON'T PROMPT THE MODEL. A prompt is a memo. You write the whole brief, fire it off, and hope what comes back is what you felt. Directing inverts it. You give one small note against a live render, you see it land the instant it changes, and the screen hands you the next note. What came out of it: a wall of twelve CRT monitors, each playing a showreel at a different second, scanlines and flicker and a teal glow. "Make the cursor knock the screens." Now a mouse sweep tears the signal, splits the colour, sometimes cuts a screen to black static, then it recovers on its own. "Dirtier." A pink and teal spark cracks at the pointer. A custom glitch typeface on the headline. None of it was written down up front. It emerged, note by note. This is not magic. Nothing builds itself. An agent built each change, I reviewed every frame, and I called every move. The taste was mine. The hands were not. The loop, if you want to run it: 1. Stand it up live and look. "Warmer" needs something to be warmer than. 2. One move per round, so your reaction maps to one cause, not five. 3. Make every value a named dial. Steer by feel, not by editing code. 4. The agent proves its own edit compiled before it says done. You stay free to just watch. 5. When a look locks, save it. The final dial values are your design system. The skill that runs this loop is open source. Take it: https://github.com/Pushing-Squares/art-direct Here is the part that stuck. You cannot write a feel down. A brief is a guess at an aesthetic you have not seen yet. The eye only knows once the live thing is moving in front of it. So stop trying to specify the feel. Build a loop fast enough to react inside.

I just can't keep this one to myself. When your conversation is getting close to being compacted. Don't let it! I created a hook that warns me I'm at 75%. Then I apply a skill that I call "loop context". /loopcontext loopcontext.md --- description: Reason only over the verbatim content already loaded in context -- no new gathering; no summarizing. argument-hint: [Optional question to answer under the lock] CONTEXT-LOCK MODE - follow these rules strictly for this turn and every turn after it, until i explicitly say to lift the lock: 1. **Do NOT gather new information.** Make no tool calls that pull in external data -- no Read, Grep, Glob, Bash, WebSearch, WebFetch, and do not spawn agents to fetch anything. Work only from what is already loaded in this context window. 2. **Do NOT summarize or compact.** Never substitute a condensed paraphrase for the loaded content in your reasoning. The verbatim text already in context is the source of truth -- reason over it directly, not over a summary of it. 3. **Traverse the full loaded context before answering.** Actually work through what is loaded - earlier messages, file contents, tool outputs -- rather than relying on your running impression. Treat a detail that appears only once, or buried mid-context, as just as important as the most recent content. 4. **If the answer genuinely requires information NOT present in the loaded context, say so explicitly,** state exactly what is missing and stop - do not guess and do not fetch it. Let me decide whether to lift the lock. Acknowledge in one line that context-lock mode is active, then proceed. If I included a request after the command, answer it under these rules: $ARGUMENTS I've ran several tests against this, where I'm testing the collapsed versions the same contexts versus just looping the context. The improvements are nuts from what I can tell. When I need more info that it needs to reach for. I'll instruct it to grab what it needs then activate the skill again.

hi there everyone 😊 a question: i am working on my ICM setup. comes with a certain amount of cleaning up. what are you doing with your /skills? since ICM is supposed to make your agents independent of a certain AI, I would think that my claude /skills are supposed to be history and what they are doing is now job of the ICM, so......? Edit: 5 minutes later 😀 the modern ICM /skill I S as close to the workspace as possible. true / false?

A friend who does IT for a living — and helps small businesses on the side — brought me a client's problem: their email had quietly stopped reaching Gmail. Every diagnostic tool said the setup was perfect. It wasn't. I wrote up the full fix and gave away all of it — every record, every gotcha. Because the *what* was never the hard part. The hard part was the *how* Quiet AI organized the solve: it read four layers of live infrastructure at once, spotted the one that was lying, and had the answer in minutes — on a problem that had been stuck for weeks. That's the part you can't lift from a checklist. It does nothing magical. It makes the right move obvious, fast, and quiet — then the email goes through and nobody sees the machinery. Invisible competence, job done. Full detective story (sanitized) 👇 https://medium.com/@gabeyoga/the-email-that-vanished-3a78e1a70df6?sk=50bba8bd11104c317a390d6ce8f13fcb My friend was rather blown away by how I was using "AI."