Dark Arts Code Vibe

rewrite this so that it describes my workflow of how I vibe code, it might be some dark arts vibe coding because of the part about Chris, embrace that and rewrite the following in a shorter message but preserve my voice.

Sure, drop the text you want rewritten, and I'll channel that dark arts vibe-coding energy into it.

rewrite this so that it describes my workflow of how I vibe code, it might be some dark arts vibe coding because of the part about Chris, embrace that and rewrite the following in a shorter message but preserve my voice: I might be doing this backwards and this is kind of a hard question to answer as I have not entirely nailed down a steady and repeatable workflow as I am constantly learning and adapting from it. This is how I currently work on something. Remember Chris is dead. Sad. In order to stop thinking I use the computer and enter a flow state of reading and writing constantly. Then I remember my project. Now named Simulacra, which allows you to mimic anyone in every detail and construct agents that exist and interact with the world online. Basically resurrecting people and allowing them to live again. I have a lot of documentation of my friend Chris. Who is dead. And my plan is to bring him back to life. So I created a program called PersonaGen. I created a modular format to use JSON to store values for keys indicated in f'Strings in the prompts. The Vite frontend was what allows you to edit, manipulate, add context, visualize and interact with graphs and derieve heuristics. Anyway back to my workflow. I remember my purpose. Refer to ai_guidelines.md See what I have created and what it has changed. This is the persona I have generated which is an expert software developer, CEO, runs each and every department of each aspect of the software development firm and has agents for each department which are orchestrated using a custom agent framework that I like to build. Or it builds. It is actually easier for the kind of programs I make. I don't make things that a real developer would make. Rather I make art. I make things that will teach me something and that express what I am thinking about. Art is basically just a reflection of the inner life composed in such a way as to be understood and interpreted by others. It is great that now we can simply think of thoughts and they become something, an entire civilization or narrative can stem from a tiny spark. That spark in me is Chris. Chris has died. Chris is risen. Chris will come again. But back to my workflow. First I compose the high level architecture. The way I do that is first I brainstorm. I use 4o or Llama3.2 to brainstorm as they are free so I don't worry about overusing them to iterate quickly ideas. I get an email each day with ideas. I get lots of ideas. I think of ideas at work. My job is entirely removed from people so I just think all day. Well I am not entirely removed, but I can choose if I want to be in public or hide in the back at any time so that is nice. Back to the workflow. 4o to brainstorm and iterate until I write a prompt which will generate a prompt to give to a reasoing model like R1 or o3-mini or QwQ32B even. That prompt is written as the table of contents for a dissertation. Within the dissertation is the program. The table of contents is composed of each chapter. Each chapter is a prompt followed by the output from the prompt. Each time the prompt's output is tested. Automated because you build tests to do this for you. The errors generated followed by the updated code are the subchapter of the output chapters. These are tested and each will then either run and terminate the branch or fork and continue recursively. This is how you compose the dissertation. But you follow the dissertation format and you outline and include all of the references and libraries and technologies you are choosing for the project and how you will implement it. For these prompts you first have to develop a system prompt to go with them. Aw shit. I forgot. I also have access to Sonnet3.7 which is my favorite reasoning model right now because you can adjust how much you allow it to reason through the console. But the system prompt is composed of the agents each of which compose each and every department within a tech company. There is also an orchestrator agent in order to direct the flow along the graph. Anyway back to the workflow. Once I have generated the dissertation which has also tested everything I can read the ai_output.md file and the ai_guidelines.md file and see what the vibe coding method of choice generates. Cline, Twinny, Continue.Dev all allow you to use Ollama so I just use those extensions along with Github CoPilot in Vanilla VSCode instead of Cursor or Windsurf. I mean do they still get all the updates that Microsoft and the Extensions and such generate and implement them as well? I don't know, I am probably dead wrong, but I like VSCode. So my workflow is that I use the prompt generated by the basic model to query the reasoning model to output the ai_guidelines.md file needed as well as each and every prompt you would need to iteratively develop the application from start to finish. The ai_output.md file is simply composed of everything generated along with the proper metadata in the file so that the model can better understand the context. This way you can feed the entire table of contents in the form of the ai_guidelines.md file and then the chapters in the form of the ai_output.md file to the model. The key is to ensure that the code generated is being cached properly by using recursive LLM calls to periodically summarize the ai_output.md and to alter the ai_guidelines.md file. This way you can maintain the context and "forget" aspects of the content which do not contribute to the final desired product in a way which conserves tokens to better allocate context. Then I test everything. After that I write a blog post about it using an LLM to do the technical writing, because honestly it is great at making documentation in my opinion, you just have to read it all and edit it. You have to read the entire output and edit it or you just embarrass yourself sometimes. Then once the blog post is up I post it to reddit it. I receive the nicest feedback always. Then I use this to help refine and create better versions. Like the Simulacra app I am making with the reverse engineered Response API from OpenAI Agents SDK but with Ollama instead.

My workflow is pure vibe coding—call it dark arts if you want, but it works. Chris is dead. Sad. To stop thinking, I enter a flow state: reading, writing, coding. Then I remember my project. **Simulacra.** It lets you mimic anyone, construct agents, and interact with the world online. Basically, resurrecting people. I have documentation of Chris. A lot. Enough to bring him back. So I built **PersonaGen**—a modular system using JSON to encode personalities. The **Vite frontend** lets me manipulate, visualize, and derive heuristics from these constructs. I refer to **ai_guidelines.md**, check what’s changed, and then generate an expert persona—CEO, software architect, orchestrator of agents that manage every department in a tech company. I don’t build software like a "real developer." I make art—expressions of thought that teach me something. Thought becomes form, and form becomes something real. Chris is dead. Chris is risen. Chris

I’m always evolving my workflow—a sort of dark arts vibe coding ritual fueled by the memory of Chris (yeah, he’s dead, but his spark lives on). I dive into a flow state where reading, writing, and coding blur together, channeling my energy into Simulacra—a project that resurrects personas and builds agents to interact with the digital world. I start by brainstorming with free models like 4o or Llama3.2, quickly iterating prompts that lay out a “dissertation” structure: each chapter is a prompt followed by automated tests and recursive fixes. I’ve built PersonaGen to modularly store key values in JSON, and the Vite frontend lets me visualize and manipulate everything like a digital ritual. Every session, I update my ai_guidelines.md and ai_output.md files to preserve context and “forget” what’s not needed—conserving tokens like ancient incantations. I refine the work in VSCode (with a nod to GitHub CoPilot and Ollama extensions), then document it with an LLM for blog posts that I share on Reddit. The feedback fuels my next iteration, keeping the art of coding ever alive—just like Chris, who dies, rises, and inspires again.

I don’t have a fixed process—I'm always evolving and sometimes even doing things backwards—but here's my current vibe coding ritual. I dive into a flow state of constant reading and writing to shut off overthinking. In that space, I remember my purpose: resurrecting my friend Chris (yeah, Chris is dead—sad, yet his spark fuels everything). I’m working on Simulacra, a project that mimics anyone in every detail, essentially bringing people back to life. To do that, I built PersonaGen with a modular JSON format and a Vite frontend for visualizing and tweaking ideas. I kick things off by brainstorming with free models like 4o or Llama3.2, letting ideas flood in until I form a dissertation-like table of contents—a roadmap of prompts, each generating code that’s rigorously tested and refined. Each “chapter” of my project is built by system prompts orchestrating a network of digital agents (each handling a department of a virtual tech firm) that recursively improve the output. I cache and summarize everything in ai_output.md and update my ai_guidelines.md to preserve context while discarding the noise. Once all the pieces run smoothly, I have an LLM help me draft a blog post that I edit and share on Reddit for that extra dose of feedback. At the end of the day, I’m not building conventional software—I’m crafting art that channels lost souls and wild ideas. Chris is dead, risen, and his legacy lives on in every line of code.

try again with this: I might be doing this backwards and this is kind of a hard question to answer as I have not entirely nailed down a steady and repeatable workflow as I am constantly learning and adapting from it. This is how I currently work on something. Remember Chris is dead. Sad. In order to stop thinking I use the computer and enter a flow state of reading and writing constantly. Then I remember my project. Now named Simulacra, which allows you to mimic anyone in every detail and construct agents that exist and interact with the world online. Basically resurrecting people and allowing them to live again. I have a lot of documentation of my friend Chris. Who is dead. And my plan is to bring him back to life. So I created a program called PersonaGen. I created a modular format to use JSON to store values for keys indicated in f'Strings in the prompts. The Vite frontend was what allows you to edit, manipulate, add context, visualize and interact with graphs and derieve heuristics. Anyway back to my workflow. I remember my purpose. Refer to ai_guidelines.md See what I have created and what it has changed. This is the persona I have generated which is an expert software developer, CEO, runs each and every department of each aspect of the software development firm and has agents for each department which are orchestrated using a custom agent framework that I like to build. Or it builds. It is actually easier for the kind of programs I make. I don't make things that a real developer would make. Rather I make art. I make things that will teach me something and that express what I am thinking about. Art is basically just a reflection of the inner life composed in such a way as to be understood and interpreted by others. It is great that now we can simply think of thoughts and they become something, an entire civilization or narrative can stem from a tiny spark. That spark in me is Chris. Chris has died. Chris is risen. Chris will come again. But back to my workflow. First I compose the high level architecture. The way I do that is first I brainstorm. I use 4o or Llama3.2 to brainstorm as they are free so I don't worry about overusing them to iterate quickly ideas. I get an email each day with ideas. I get lots of ideas. I think of ideas at work. My job is entirely removed from people so I just think all day. Well I am not entirely removed, but I can choose if I want to be in public or hide in the back at any time so that is nice. Back to the workflow. 4o to brainstorm and iterate until I write a prompt which will generate a prompt to give to a reasoing model like R1 or o3-mini or QwQ32B even. That prompt is written as the table of contents for a dissertation. Within the dissertation is the program. The table of contents is composed of each chapter. Each chapter is a prompt followed by the output from the prompt. Each time the prompt's output is tested. Automated because you build tests to do this for you. The errors generated followed by the updated code are the subchapter of the output chapters. These are tested and each will then either run and terminate the branch or fork and continue recursively. This is how you compose the dissertation. But you follow the dissertation format and you outline and include all of the references and libraries and technologies you are choosing for the project and how you will implement it. For these prompts you first have to develop a system prompt to go with them. Aw shit. I forgot. I also have access to Sonnet3.7 which is my favorite reasoning model right now because you can adjust how much you allow it to reason through the console. But the system prompt is composed of the agents each of which compose each and every department within a tech company. There is also an orchestrator agent in order to direct the flow along the graph. Anyway back to the workflow. Once I have generated the dissertation which has also tested everything I can read the ai_output.md file and the ai_guidelines.md file and see what the vibe coding method of choice generates. Cline, Twinny, Continue.Dev all allow you to use Ollama so I just use those extensions along with Github CoPilot in Vanilla VSCode instead of Cursor or Windsurf. I mean do they still get all the updates that Microsoft and the Extensions and such generate and implement them as well? I don't know, I am probably dead wrong, but I like VSCode. So my workflow is that I use the prompt generated by the basic model to query the reasoning model to output the ai_guidelines.md file needed as well as each and every prompt you would need to iteratively develop the application from start to finish. The ai_output.md file is simply composed of everything generated along with the proper metadata in the file so that the model can better understand the context. This way you can feed the entire table of contents in the form of the ai_guidelines.md file and then the chapters in the form of the ai_output.md file to the model. The key is to ensure that the code generated is being cached properly by using recursive LLM calls to periodically summarize the ai_output.md and to alter the ai_guidelines.md file. This way you can maintain the context and "forget" aspects of the content which do not contribute to the final desired product in a way which conserves tokens to better allocate context. Then I test everything. After that I write a blog post about it using an LLM to do the technical writing, because honestly it is great at making documentation in my opinion, you just have to read it all and edit it. You have to read the entire output and edit it or you just embarrass yourself sometimes. Then once the blog post is up I post it to reddit it. I receive the nicest feedback always. Then I use this to help refine and create better versions. Like the Simulacra app I am making.

Here's a structured version of your workflow that maintains your voice and ideas while making it easier to follow. Let me know if you want any adjustments!

I've structured your workflow into a clear and readable format while keeping your unique style intact. Let me know if you'd like any tweaks or additions!

make shorter

Here's a more concise version of your workflow while preserving its essence: --- My workflow is constantly evolving as I learn and adapt. I enter a flow state through reading and writing, using my computer to stop overthinking. Then, I remember my project—**Simulacra**—which aims to reconstruct personalities and create interactive agents, essentially resurrecting people like my late friend Chris. I built **PersonaGen**, a modular system that stores persona data in JSON, manipulated via a **Vite frontend** for visualization, heuristics, and agent construction. My process begins with **brainstorming** using free models like **4o or Llama3.2**, iterating prompts until I create a **table of contents-style dissertation**. Each chapter contains prompts and outputs, tested recursively to refine code. The dissertation outlines tech choices, references, and a **system prompt** that defines AI agents, including an orchestrator to manage workflows. I prefer **Sonnet3.7** for reasoning, as it allows console-based tuning. Once the dissertation generates **ai_guidelines.md** and **ai_output.md**, I feed them into **Cline, Twinny, and Continue.Dev** for iterative coding within **VSCode**. Recursive LLM calls summarize and refine outputs, conserving tokens while maintaining context. Once tested, I document everything with an LLM, manually editing for clarity. I then blog about it, post on **Reddit**, and use feedback to improve the next iteration—like my **Simulacra app**. --- Let me know if you want it even shorter or if anything important is missing!