You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We now have one key per builder (#1404: merged Builder Action, phase-aware press that also selects the builder). But covering a builder end-to-end needs several actions (approve a gate, run dev, send review feedback), and the recently added queued code-review feedback has no hardware trigger. We want an efficient SD+ layout where the deck is a remote control and VSCode is the screen, bound by a single shared selection.
Row 2 (4 keys): action palette, fixed in place, always acting on the selected builder: [Approve gate] [Run Dev] [Send Fb (N)] [ Next / free ]. Rejected: activating a builder to re-flow all keys (modal, kills muscle memory, ambiguous state).
Dials collect, key commits. Repoint each diff dial press from immediate send to submitting the chunk as feedback via a mode-neutral verb (feedback-file / feedback-hunk / feedback-selection). VSCode routes it forward-now or enqueue per the workspace setting. Dial rotate = navigate (unchanged), tap = jump-to-first (unchanged).
Send Fb (N) key = flush. Fires a send-queue verb; badge N mirrors the per-builder queued count from the overview. In immediate mode N stays 0 and the key is inert; in queued mode N climbs and the key sends. No deck-side mode inference.
Legibility, not just correctness. Because the same dial press sends in one workspace and queues in another, surface the current mode on the dial touchscreen (e.g. Files · queue vs Files · send) so a press is never a surprise.
Overview wire. Add a per-builder queued-feedback count so the deck can render N (alongside the queue capability already added on the VSCode side).
No approval semantics leak. Row 2 Approve acts on the selected builder and routes through the existing confirmation. Decide deliberately whether it supersedes the standalone ApproveGate singleton, or the singleton's "jump to next pending" folds into a Row 2 Next / attention key (jump selection to the highest-priority builder). Don't ship two muddy approve affordances.
Verb behavior change (forward-* immediate to feedback-* mode-routed + send-queue) touches the VSCode command relay; coordinate with the review-queue work.
Suggested protocol
PIR: the dial feedback loop, the mode-follows-setting behavior, and the Row 2 palette all need hardware verification before a PR.
Problem
We now have one key per builder (#1404: merged Builder Action, phase-aware press that also selects the builder). But covering a builder end-to-end needs several actions (approve a gate, run dev, send review feedback), and the recently added queued code-review feedback has no hardware trigger. We want an efficient SD+ layout where the deck is a remote control and VSCode is the screen, bound by a single shared selection.
Ground truth
store.cursor,selectedBuilder());DevServerAction, the diff dials,ScrollNav, andZoomNavact on it. After Stream Deck: merge Fleet Slot into Builder Action (phase-aware press) #1404, pressing a Builder Action key sets that cursor (syncToBuilder).forward-*, tap = jump-to-first). They are at gesture budget.Requirements
feedback-file/feedback-hunk/feedback-selection). VSCode routes it forward-now or enqueue per the workspace setting. Dial rotate = navigate (unchanged), tap = jump-to-first (unchanged).send-queueverb; badgeNmirrors the per-builder queued count from the overview. In immediate modeNstays 0 and the key is inert; in queued modeNclimbs and the key sends. No deck-side mode inference.Files · queuevsFiles · send) so a press is never a surprise.N(alongside the queue capability already added on the VSCode side).ApproveGatesingleton, or the singleton's "jump to next pending" folds into a Row 2 Next / attention key (jump selection to the highest-priority builder). Don't ship two muddy approve affordances.Out of scope / relation
forward-*immediate tofeedback-*mode-routed +send-queue) touches the VSCode command relay; coordinate with the review-queue work.Suggested protocol
PIR: the dial feedback loop, the mode-follows-setting behavior, and the Row 2 palette all need hardware verification before a PR.