brain-autolearnlisted

# Brain Autolearn — Turn Repetition Into Leverage ## Overview A mistake written to the vault is not a lesson learned; a snippet retyped four times is not a tool. This skill closes that gap: it scans the session for **repetition of two kinds** — repeated mistakes and repeated actions — and converts each into an artifact that fires by itself next time. Two principles decide everything below: - **Rule of three.** One occurrence is noise. Two is a candidate — count it. Three (within a session, or across sessions via the inbox) is a signal — act. Exception: a cheap deterministic script pays for itself at two — build it at ×2. Rules, skills, and anything the owner must review wait for three. - **An artifact nobody recalls is a second vault of forgotten things.** Every artifact this skill creates gets a trigger in tool-routing so it surfaces when needed. Typical yield: 0–2 artifacts per session. A clean session must end in seconds with "nothing to learn" — a heavy pass will be skipped, and a skipped pass learns nothing. ## Step 1: Repetition sweep Ask each question against THIS session. Note only concrete moments. 1. Did I make the same mistake twice — or once here and it is already a `mistake` note in the vault? (verify with a targeted `brain_search`) 2. Did the user correct me the same way twice? Corrections are first-class signals, not just memory material. 3. Did I write the same snippet or command sequence three times, with variations? 4. Do 3+ tool calls