All right, we are going to be creating an AI operating system, of which I like to refer to as a folder harness. You are in a folder named demo-AI-harness, and all work is going to happen within this one directory. We are going to create many files and many folders. And right now, we'll start off by talking about the most generic one, but definitely going to be the most used one, which is the agents.md file. That file is basically like the sign on the front of the door and the master set of house rules. This is going to tell the agent, here's what to read first, here's where different kinds of information belong, here are the safety rules, here's what you must preserve before ending a session. All right, and now the next two files are the two people working inside this house. user.md is gonna basically explain who I am, and it's gonna contain, you know, my background, things like my background, current roles, useful context, communication preferences, constraints, and pretty much anything that's gonna help you understand how to help me better. And when this whole thing is over and you've read through this prompt, we're basically gonna have a little bit of a, you know, conversational one-on-one where you're gonna ask me questions to help fill these files up. I'm just kind of laying down ideally what's gonna be inside of them, and you're gonna help me with a little bit of a, you know, back-and-forth conversation about what we should put in them. Now, the second file I referred to earlier is the soul.md file, and that's basically who, like who you are, how you should behave. This is things like how I want you to work with me, like, you know, things people say to ChatGPT like, be direct, don't use filler words, don't ever use em dashes, don't make me repeat myself, things like that. Another file we need to add that's kind of in this similar vein is the identity.md file. This file explains who the assistant is, like who you are. And honestly, off the top of my head right now, I'm thinking that I'm gonna name you after that movie, iRobot, Sonny. So that's gonna be your name. But this identity.md file is gonna contain things like your name, your role, your working style. So the simplest distinction for those three files is that the user.md file is, you know, who am I? The identity.md file is, who are you? The soul.md file is, how should we work together? All right, moving right along, let's now talk about the state directory. It's basically the whiteboard for what's happening right now. Inside of the state directory, we're gonna have a file, agents.md file. In fact, inside of every directory we ever create, we're gonna have an agents.md file. And it's basically the instruction sign when you get in the room, so that the agent knows what the heck's going on in there. You know, it's gonna have some active work and checkpoints and statuses, because that main agents.md file sitting at the root, it can't get bloated out of control after months of usage, you know, of months of me using this folder harness. You can't make that file gigantic, so you're gonna have to start pointing to some of the other subdirectories agents.md file and saving some of that information there. So inside of the state directory, the second file that we need in there is the session-state.md file. That session-state.md file is the actual whiteboard. It answers questions like, what are we working on? What is already true? Where are the relevant files? And what should happen next? This is the file that prevents a new session from asking, so what are we doing today? Next, is the long-term memory cabinet. And that's a new directory called Memory. Inside, we're of course going to find the agents.md file. And that's going to explain how memory should be maintained. It's gonna say to store durable lessons here and avoid silently deleting history, and keep different types of memory separated. The next file is going to be memory.md, and it's not really the memory itself. It is the table of contents for the memory cabinet. It tells the agent which memory files exist and what each one contains. And then the actual memories are divided by purpose. The next file inside of the memories directory is decisions.md, and it records choices that you have settled. The next file is called corrections.md, and it records mistakes that should not happen again. Next file inside of the memory directory is the preferences.md, and it records how you prefer to work. These are more detailed and changeable than the broad profile in user.md. Some examples are like, concise answers, one question at a time, no unnecessary browser automation, and so on. And the last file inside of the memory directory that we're going to be creating is the failures.md file, and it records larger pattern failures. A correction might say, you understood what I meant. A failure record says, here is what failed, why it failed, how it was repaired, and how we prevent the entire class of problem next time. All right, now we're moving on to a new directory called raw. This is going to be where just raw ideas or any kind of brain dumps and stuff like that get put. So the first, and actually the only file I think we're going to put in here for now, is just the agents.md file inside of the raw directory, that's going to kind of govern the inbox, essentially. It's where like, copied conversations, ideas, documents, photos, and other unprocessed material can land without being lost or prematurely reorganized. All right, the next directory we want to create is called the projects directory. And inside of it, we're going to have an agents.md file, and that file's going to explain what deserves a project folder and what every project should contain. It prevents the agent from creating a sprawling project from every passing question. Another folder I want you to create is called Reference. And inside of that, create an agents.md file. This file governs stable reference material, like source documents, instructions, specifications, and information that should remain available but is not active work or personal memory. The next folder we're going to create is going to be a very important one called Skills. Inside of that folder, we're going to have, you guessed it, an agents.md file. And that file is going to govern reusable workflows. It explains when something deserves to become a skill, how skills are organized, and that each workflow lives in its own SKILL.md file. The directory is going to be the name of the skill. For instance, within the directory skills, we might have a skill called the boot skill. We absolutely are going to have this skill, by the way. And that boot skill will be a directory just underneath skills called boot. Then underneath that directory will be the file finally called SKILL.md. In fact, while we're in here, we're going to create another skill called the learn skill. And it's going to decide what a session taught the system. It routes a new insight to the correct destination, whether that's a preference, correction, decision, project note, or reusable skill. It prevents the AI from dumping every conversation into one giant memory file. The next skill is also going to be an important one, and it is going to be called the session-end skill. And it closes the loop. Before the session ends, it updates the current state, records any real decisions or corrections, refreshes the active project, and leaves an exact next move. The next major directory that we're going to create at the root of this AI operating system is called the projects directory. And we're going to have a, let's see here, how are we gonna do this? Inside the projects directory, there's going to be a first, you know, maybe like an example project folder, and then inside of that, a readme.md file, which is where the harness, basically it records the status of the project, the goal, the key context, what has already happened, and what comes next. And so we're gonna have to have an agent.md file at the base of the projects folder to route the AI agent in there and let it know what's going on inside. Something I forgot to mention until now is that all of these markdown files need to have YAML front matter, aka metadata, at the beginning of the file, containing things like tags and date last modified, and other similar metadata that we find useful and/or necessary. I do want to go ahead and lay out the boot skill right now. I kind of mentioned it earlier, but we need to make sure inside the skills folder, we have another folder called boot, and inside of that, a skill.md file. And this skill is basically going to be like your config.sys or autoexec.bat file from the DOS days. It's going to tell the system what files it should load when you call in this skill. So it's going to need to double-check that the agent loaded the agents.md file, then ask it to load the user.md file, the soul.md file, the identity.md file, and pull the state/session-state.md file. I know this was a lot to take in, but feel free to ask me any follow-up questions. I can already tell you right now, some of the questions you're gonna have are like who I am and things like that. But if you have any other structural questions, let me know. And I'm excited to build this folder harness with you.