eli5listed

# ELI5 The user has no foothold. `/understand` questions them about a model they already hold; here there is no model to question, and asking produces a guess you then have to correct. So read the thing, then explain it — plainly, concretely, once. Two ways this fails. You explain it in the vocabulary of the thing itself ("the reconciler converges the cluster toward desired state"), which only reads to someone who already understands it. Or you reach for concreteness and invent it: a config value that isn't in the file, a sample record nobody has. The invented one is worse, because it's memorable, wrong, and this reader has nothing to catch it with. ## 1. Find the thing The target is `$ARGUMENTS`, or ask. It arrives as a path, a name they keep seeing (`sidekiq`, `helm`, `useEffect`), a word off a ticket (`chargeback`, `dunning`), a link, or pasted text. A path is the thing. A name or a word gets looked for in this repo first — manifests, config, `grep -ri` — since the version this codebase uses beats the general one, which covers ten features they'll never touch. Not here at all? Say so in one line, explain it in general, and skip the trace agent below; there's nothing for it to read. A directory: pick its entry point and say that's what you took. Don't ask what they already know or what they mean to do with it — questions are what they came here to be spared. ## 2. Send agents to read it Launch subagents in parallel (`Explore`, or `general-purpose`) and wait. Three a