brownfieldlisted

# Brownfield `stack-profile` opens with *"Generate, do not author"* — run the scaffolder, let the tool produce correct config. That is right for an empty directory and wrong for everything else. On a project that already exists there is nothing to generate. There is something to **discover**, and the difference matters because a profile written from assumption is a contract every other agent then obeys. So the rule here is the mirror of that one: **discover, do not assume.** Every line of the profile traces to a file you read or a command you ran. ## 1. Get the map before reading anything ``` node <plugin>/scripts/survey.mjs . ``` One command, under a second on a large repository. It reports where the code actually is, the test ratio, the entry points, which files the last four hundred commits touched, what is generated or vendored, and — the part to read carefully — **what it could not establish.** Churn is the most useful line in it. Directory size tells you where code accumulated; churn tells you where the work is, which is almost always where your change lands. ## 2. Read with a question, and record what you did not read You cannot read ten thousand files and you must not pretend otherwise. Senior practice is to decide what to understand and what to skip, with a goal — not to front-load the whole system. Follow one thread end to end: pick the entry point nearest your task and read outward until you can say what happens when a user does the thing you are about to