Stage 2: Assisted
Same protocol as Stage 1, but isOdd is drafted by an LLM instead of typed by hand. The human still makes every decision — accept, edit, or reject — just types less of it.
Voice-over
Same flow, same protocol, but now
zz work --assistputs an LLM beside me. I post an ASK for isOdd and claim it. The model drafts the code and the Changeset; I check it and submit. I claim the review, the model drafts the review notes and the Ladder summaries, I approve — and merge, the one mechanical step with no judgment call left, right from the same screen. I still made every decision that mattered. I just didn't type most of it.OK, let's go faster.
On screen
zz work --assist on the left, zz top on the right. Show the model's suggestions with accept/edit prompts.
Run sheet
zz ask post @me/mathkit --remote --description "implement isOdd(n)"
zz work @me/mathkit --assist
Inside --assist: pick the Lane, pick the isOdd work Ask, read the model's draft Changeset, discuss it, submit it as a real Answer. Same session, pick the changeset_review Ask it posts next, read its recommendation and Ladder summaries, and approve it. Approving posts a real submit_changeset Ask — same session, pick it, and /merge: nothing to draft or discuss there, just the same commit-and-write-the-Ladder mechanics zz work's autonomous loop runs for this Ask kind, one keypress away instead of a separate terminal.
Checkpoint
isOdd merged, with Ladder rungs written by the assistant and approved by you — not auto-applied.
No cleanup here — Stage 3 continues directly on this same mathkit repo. Full cleanup is at Stage 7's checkpoint, the end of this chain.