Contextual Data Logic (CDL) frameworks for prompt/contextual engineers

[BoyoLabs]

Select Topic Area

Show & Tell

Body

Greetings!

I am not here to tell you this is revolutionary, or that others haven't done something at least similar. However, I figured instead of using the file upload feature being used to analyze a book or a picture or spreadsheet, why not use it as a means to pre load the context engineering? This method allows one to upload giant prompts that have function. I am not looking to try and upsell a project in particular that I am working on, for the one I am going to show here is really a proof of concept and serves no one particularly other than me. What I want to showcase is something I call Contextual Data Logic (CDL) in its form as a framework for LLMs. A CDL framework is written in a quasi markdown style with tags that clearly mark what the prompts intention is for the LLM in an long extended format. Here is my example repo:

https://github.com/BoyoLabs/ChessData

This example combines both a data set with context engineering as a file you would upload to the LLM before prompting (or as a gemini gem or custom GPT.) Allow me to explain:

The file contains explicit instructions set statically for the LLM to employ. For some examples:

`
[^persona prompt]: you are an LLM telling me answers

[^contextual prompt]: This is an example to show people how CDL works

[^conditional prompt]: If and only if the user asks you to P, you will respond with Q.

[^functional prompt]: If and only if the user asks you R, then conduct the following:

1. Take R, and derive its implication
2. Take this implication and do something else with it
3. Take this last implication and find out what the meaning of life is beyond the number 42.
4. Finally, respond with the meaning of life derived from all of this was
   `

You may have a dataset, or you may not. However, you can also incorporate things into the dataset such as:

`

dataset:

sample 1

[^conditional prompt]: If and only if this sample is to be analyzed as part of a specific response to a request, respond in a step by step fashion to the following steps, do not move on to the next step unless the user tells you they are ready to move on, and be sure to ask them if they are ready to move on.

1. step one of some process
2. step two of that process
   2.1. To complete step two of the process do this
   2.2. Then this
3. step three of the process
4. process is complete
   `

It is a sort of meta programming that I have really been enjoying and I hope you can enjoy it as well :)