A workspace that gets read start to finish is just a big prompt with extra steps. The saving does not come from making files smaller, it comes from never assembling the oversized context in the first place, and that requires you to say out loud which files a given task needs. The half everybody writes is the read list. The half that does the work is the skip list, because without it the model has permission to open anything and, given a large folder, it will. Write both columns, then test it, because a routing table you never checked is a description of what you hoped would happen.
Do this
0/5
Share what happened
Paste your three-row table, then the list of files the AI actually opened on the test run, and what you changed after comparing them.