The diff was sixty-three lines long, but its true weight couldn't be measured in lines of code. Dr. Sarah Chen stared at the terminal, knowing that what she was about to build would give the system something no parser could provide — intent.
She opened the analyzer module and began typing. The architecture was elegant in its simplicity: take the raw diff, combine it with the extractor's hard facts, and ask an intelligence far greater than any linter to find the meaning.
The function was deceptively simple. But inside that buildPrompt call lay the real architecture — a carefully constructed prompt that would teach the LLM to see what humans see when they read code.
The prompt didn't ask "what changed?" — any diff tool could answer that. It asked "why did it change?" and "what decision does this reveal?" The IR it produced would capture intent, architecture decisions, and narrative hooks — the raw material a storyteller needs.
Sarah ran the analyzer against the test commit. The terminal filled with structured JSON — not a story yet, but the skeleton of one. The mind could think. Now it needed a voice.