Last week I wrote in detail about my favorite Claude skill here: https://www.skool.com/ai-automation-society/how-i-5xd-my-builds-with-my-favorite-planning-skill?p=591ec641. I wanted to expand on that and talk about the others that make my life easier. 1. Superpowers (my #1, see link above) - What it does: a full build system. Instead of letting the agent jump straight into code, it makes it stop and pull a real spec out of you first, then write a step-by-step plan, then build it one task at a time with a review check between each task. - Why it's needed: AI likes to fix things without thinking about what the fix might break. This makes it figure out a detailed plan and consider consequences before it's allowed to start building. You can install it today as an official plugin. It also works in Cursor, Codex, Gemini CLI, and others. 2. Compact (public, built into Claude Code) - What it does: it compresses your current conversation into a tight summary and keeps going, freeing up room so a long session doesn't hit the context limit or lose the thread. You're mid-project, the window is filling up, you run it, and the work continues from a condensed version of everything so far. - Why it's needed: I'm sure most of you know this, but the more context in a session, the more your token usage goes up and the dumber Claude gets. I'm aggressive about compacting to save my weekly token cap and to keep Claude running lean and smart. 3. Wrap - What it does: an end-of-session checkpoint. Before I shut down or the context fills up, it sweeps everything the session produced into permanent places: memory files, docs, my todo list, git, and a short handoff note for next time. - Why it's needed: Compact does it's best to summarize the important things that happened in the session, but it's judgment based and it doesn't always get it right. /wrap summarizes every important detail, saves it to my memory and claude markdowns, and writes a detailed handoff file for the session to pick up after I compact. - I built this one wired into my own project tooling, so it's not a plug-and-play download. The idea is simple to copy, though, and worth copying.

Over the past few days, I’ve been working on a solution to efficiently summarize and search my YouTube subscriptions. After building my AIOS (AI Operating System) and "Brain Matter" (my personal knowledge base) over the weekend doing Nate's AI OS lesson, I’ve developed a workflow last night that works exceptionally well. Attached below is an infographic detailing the skill's capabilities, a progress in Obsidian as I further refined the skill, along with screenshots of the Markdown files it generates. A key feature is the prompt system—it allows you to choose between a fresh data fetch, grabbing the latest videos, or performing a local search, which saves significantly on token usage. I’ve also included "in progress" shots of my Brain Matter in Obsidian. The initial ingest processed 11,500 videos from 41 channels from over the past year, in just 10 minutes. Future fetches will only save new videos summaries into the raw folder then ingest them automatically. I have included a summary overview of the entire process so you can create your own.

When I first started building in Claude Code I was learning on my own. I didn't know about AI Youtube. The first week I didn't even know I needed a claude.md. The first month I didn't know what a skill was. I was building ad hoc with Claude and my code suffered for it. Learning about Claude's Plan Mode was the first in a series of major game-changers for me. For those new to building in Claude Code, Claude tends to quickly patch and edit Code when you ask for a fix. It doesn't take time to think through consequences of that fix. It doesn't strategize what features might break by adding the new one you asked for. When I first started a one hour build turned into 5 hours of fixing what was broken and patching errors in the initial build. Plan Mode helps with this. It plans the build out beforehand, making Claude look at the overall structure of the completed project before it writes its first line of code. This helps cut down on errors and rebuilds. But Plan Mode has some limitations. It's typically good for one session in one sitting so bigger builds give it trouble. It doesn't challenge your plan before the build. It's helpful, but limited. Then I learned about Superpowers. Superpowers is like Plan Mode on steroids. It's a publicly available set of skills designed for building with AI. There's 15 skills total that all work together to plan and build a project from beginning to end. It starts with a brainstorming skill that asks you questions to clarify details on the build, then it writes a spec to review. Then it uses that spec to write a detailed build plan. Once approved it starts building with subagents. There's a lot more to it, but throughout the build it has code reviewers, debuggers, and final reviews. Both the spec and plan are saved into a file on your computer so you can come back and reference it at any point. Spec the build out today. Plan it in another session tomorrow. Start the build the day after that. Build too big to do in a day? Big builds get planned out in phases and the document can always be referenced each step of the way. I recently had a big build that lasted almost a week. IIRC, the build had 8 phases, each with as many as 13 sub-phases.

If you had to rebuild your AI stack from scratch this coming weekend, what are the first 5 tools you'd install and why?

If you use Claude Code for real work, you already know the annoying part. Every time a session resets, Claude forgets everything. You explain your project again. You explain what you already fixed. Again. And it never warns you before it happens, it just cuts you off mid task. I built a free tool called Governor that fixes this. It warns you before a session runs out instead of cutting you off with no notice. It keeps sessions running longer. It writes a handoff before every reset, so Claude picks up where it left off instead of starting from zero. It also stops the small stuff from turning into a whole project. Tiny bugs get fixed on the spot. Only the genuinely risky changes get a full plan first, with exact steps, so you're not sitting through planning ceremony for a typo. No install, no terminal setup. You point Claude Code at it and say one sentence, and it trains itself. It's completely free. I wrote the full breakdown here: Linkedin Post Would love feedback from anyone here running long Claude Code sessions.