I'm excited to share that my "Temporal Memory Bank" prompt was just accepted into the official Cline prompts library. I'm incredibly proud of this because it's been a game-changer for how I work with AI, and I hope others find it useful too.
The main idea is to give the AI a deliberate speed bump. I found that even powerful models like Claude 3.7 Sonnet can rush ahead and try to do too much at once, leading to messy results. My prompt is designed to slow things down and enforce a philosophy I call "Baby Steps™".
So, what are Baby Steps™? It's a simple idea: make the smallest possible meaningful change at any one time. Instead of telling the AI "build me a website," I use the memory bank to focus on one, single, well-defined task from a list of known issues.
My workflow now looks like this:
1. __Plan Mode:__ I have the AI read the memory bank and craft a detailed, step-by-step plan to tackle just one issue. We go back and forth until the plan is solid.
2. __Act Mode:__ Once the plan is set, I switch to a more agentic model to execute the steps. The AI's focus is narrow and the goal is clear, which leads to much better outcomes.
It's a way to force a more methodical, human-like approach to problem-solving, and it's made my collaboration with AI far more effective.
Obligatory blog post: https://convergence.ninja/post/blogs/000010-A-Better-Memory-...
I'm excited to share that my "Temporal Memory Bank" prompt was just accepted into the official Cline prompts library. I'm incredibly proud of this because it's been a game-changer for how I work with AI, and I hope others find it useful too.
The main idea is to give the AI a deliberate speed bump. I found that even powerful models like Claude 3.7 Sonnet can rush ahead and try to do too much at once, leading to messy results. My prompt is designed to slow things down and enforce a philosophy I call "Baby Steps™".
So, what are Baby Steps™? It's a simple idea: make the smallest possible meaningful change at any one time. Instead of telling the AI "build me a website," I use the memory bank to focus on one, single, well-defined task from a list of known issues.
My workflow now looks like this:
1. __Plan Mode:__ I have the AI read the memory bank and craft a detailed, step-by-step plan to tackle just one issue. We go back and forth until the plan is solid. 2. __Act Mode:__ Once the plan is set, I switch to a more agentic model to execute the steps. The AI's focus is narrow and the goal is clear, which leads to much better outcomes.
It's a way to force a more methodical, human-like approach to problem-solving, and it's made my collaboration with AI far more effective.
A huge shout-out to Nick Baumann ([](https://github.com/nickbaumann98)<https://github.com/nickbaumann98>) for his awesome original work on the memory bank concept, which was the foundation for about 90% of this prompt.
Editor note: no markdown in posts, just links
Direct link to Nick's temporal memory bank prompt: https://github.com/cline/prompts/blob/main/.clinerules/tempo...
Cool to see, love this formalism/breakdown. I feel like I have a lot to gain with these structures, these different component markdowns.