spcl/graph-of-thoughts ? reverse-engineered prompt

Build me a small Python framework that helps me solve tricky problems with a large language model by breaking the work into a graph of steps instead of just one long prompt.

I want it to let me define a sequence of operations like generating ideas, scoring them, and checking results, then run that flow with an LLM and save the whole thought process to a file so I can inspect it later. Please include a simple way to plug in different prompt templates and parsers, and make it flexible enough to model simpler approaches too, like chain of thought or tree of thought.

Also include a couple of working examples I can run from the command line, like sorting a list of numbers and counting keywords. Make it easy to install, easy to extend, and clearly documented. If you need to, look up current docs online for the LLM setup details.