You all know the drill: you want to write prose using an AI because it can draft faster than you can type your ideas out. Next thing that happens, you find yourself in an editing loop: change this, fix there, explain your world in more detail.

The chat window scroll bar shrinks to 2 pixels and the chapter gets worse every turn. You wonder if your prompts got bad or if the AI has forgotten your world.

What actually happens is the context window starts working against you. A context window works fine if you keep talking about ever-changing topics. When you write prose and ask for endless changes, the context fills up with copies of the same text in slightly different colors. Your actual changes become noise, not focus.

My solution: a clean chat with zero context to fix one piece in one scene.

Instead of spilling all my chapters into the prompt, I split my world setup into small Markdown files. When I start a new chat, all I put into context are mentions of those world files, the actual chapter, and relevant cast sheets. I let the agent do the heavy lifting. That's what Cursor is best for: finding its way around tons of files and making a code change like a surgeon.

Side effect: your entire book stays as local files on your disk. My tool does not take your work hostage.

When you ask for a change, you don't get back a full rewrite in a browser window—you get a targeted DIFF view on a text file. The chat context stays crisp and focused on exactly what you ask.

So the paradigm change I promote is: go away from long-term chat sessions. Forget about "the next prompt will fix it." Start working in a controlled environment where rules are fetched on demand. You do not prompt anymore—you have a creative meeting.

Since I don't trust anyone, not even myself, I added two mean critics to the system: 1. The AI Slop Critic: looks for where something sounds like generic AI prose. 2. The Story Critic: finds contradictions or plot holes in your story.

The first time I ran the Slop Critic on a chapter, it came back and crushed my heart. It read so bad! But it gave a "fixable" verdict. To fix it, you right-click on the critic's finding and start a chat with your writer. A clean chat opens with the relevant context and the brief inside. All you do is hit return.

Because these are plain text files, it also includes Git integration. It commits the chapter state upfront before each agent edit, showing you a diff to all historical versions with a few clicks.

If you're still here, I will finally give you a name and a link to play with:

Buckram Studio: https://open-vsx.org/extension/buckram/buckram-studio Or search for "Buckram Studio" inside Cursor. It's freemium so you can test the writing/editing process up to 50k words.

Post on X: https://x.com/TorrenBramley/status/2079210475066114326

Ask me any question you want, or crush me like my mean critics :D

ChatBOT chapter thread is two weeks old. That's why the prose went soft