I've been working on an agent that seems to be reading all the documents twice. I'm suspicious that this is because it's reading the documents automatically because they are tagged in the input form, and then again in the read steps where I assumed they would actually be read in.
Is there a best practice for reading in documents in agents where documents are explicitly tagged? Want to make sure I'm building this as efficiently as possible and not unnecessarily using tokens or context window.
You can see the same 5 documents are read in first altogether, then separately again (they are split between two different read steps so the can be referenced separately).
In debug mode, it only shows them being read in one time though (the five docs are split across the two steps and are not repeated, verified through the debug input/output):
