Maybe, I want to ask the LLM to work as a specific procegure, but the long Promt may lead the fault, if I can invite a "workflow" to confirm the LLM will work in proper manner that if they do somethings wrong the Harness can feedback and stop to execute the tool calls, there are much less mistakes.Perhaps...
Yes perhaps in this case you should watch the video of Garry Than about Personal AGI.
He give a really interesting approach about creating your own context each days.