APPENDIX H: RESUME PORTFOLIO BUILDING CAPSTONE PROJECTS

PROJECT OVERVIEW Picture a student writing a report at a small desk. The desk is only so big, so the student cannot spread out every book at once. A sensible student reads one book, writes the important fact on an index card, and then puts the book away to make room. If the student gets interrupted and comes back the next day, the index cards tell them exactly where they left off. Project 2 builds that sensible student. Your agent will answer a question from six invented garden documents, reading one document at a time, saving the key fact from each in a notes file, and folding old conversation into a short summary whenever the desk, which means the context window, gets too full. The context window is the limited amount of text a model can look at in one turn, and compaction is the act of replacing a long stretch of old conversation with a short summary. You will also prove that the agent can be stopped halfway, started again, and carry on from its notes without rereading what it already finished. The project relates to Chapters 18 through 22 and to Chapter 66, and to the second stage of the Practice Lab, which also added a context builder, compaction, and a progress notes file. It is different from the lab because the documents are invented garden pages with one planted fact each, and because you will measure compaction and restart behavior with tests. It reuses two files from Project 1, which are gate.py and tools.py, unchanged. Use only the invented documents that the project creates for you, or other invented text, and never real personal data.