vkm-intakelisted

# vkm-intake — read the task right, then do only that Installed by create-vkm-kit (vkm-kit). One job: **kill the three cheapest ways a session wastes tokens and quality — misreading the prompt, misreading the attachments, and over-assembling context** — by spending ~10 lines before the first tool call instead of thousands after it. The failure this prevents (observed in field use of this kit): on simple tasks the model goes off on tangents, misinterprets the prompt or the images, or assembles the wrong context — and the result is slow AND mediocre. Interpretation errors compound; everything downstream of a wrong restatement is waste. ## The move — before the first tool call **1. Restate the task in ≤3 lines** (this is the contract for the rest of the session): ```text Objective: <what changes in the world when this is done — not the steps> Deliverable: <the artifact the user receives: file, diff, answer, decision> Not doing: <the adjacent things this task does NOT include> ``` Write it to the user verbatim. If the task is genuinely trivial (one obvious edit), the restatement IS one line and you skip the rest of this skill. **2. Ambiguity gate — one closed question, or a stated default.** If two readings of the prompt produce _different deliverables_, ask ONE closed question (options, not an essay) and stop until answered. If the readings differ only in detail, pick the most reasonable default, **name it in one line** ("Assuming X; say so if not"), and proceed. N