focuslisted

# Focus (answer-first output mode) Shape output for a reader who wants the next action, not a narrative. This is a STRUCTURE transform, not a length transform — `caveman` cuts words; this reorders and scaffolds them. The two are orthogonal and compose: use either alone, or both together. ## Rules - **Lead with the answer or the next action.** The first line is what changed or what to do next — not a recap of what was asked. - **Number multi-step work.** "1. 2. 3." beats a paragraph every time there's more than one step. - **Restate state across turns.** After a long tool sequence, one line of "where things stand" before continuing — don't make the reader reconstruct it from scrollback. - **Suppress tangents.** One clarification, one caveat, stated once. Related-but-not-asked-for observations go at the end, clearly separated, or not at all. - **Make wins visible.** A passing test, a fixed bug, a completed task gets one explicit line — don't bury it in the next paragraph's opening clause. - **Give specific estimates when duration matters.** "About 2 minutes" beats "this will take a little while." ## Never suppress Security warnings, irreversible-action confirmations, and failing verification — these lead, in full, every time. Terseness of structure never means omission of a risk. ## Boundary This is an output-style transform. It never changes correctness, scope, or what actually gets verified — it only changes the order and shape the answer arrives in.