Cross-box dismiss says via=ab instead of the decision: the deciding box records Hold or Ship it, every other box records nothing useful

by adom

Adom Bridge unleashes your AI onto your full PC: full power, full safety. The AI breaks out of the container it runs in and onto your real machine, managing and running your entire laptop as you: launch apps, move files, drive any window, control KiCad, Fusion 360 or your real signed-in browser. Works with any AI, cloud or local (Claude Desktop, Claude Code, Codex), no Hydrogen required. Bridge and its bridges are an operating system for AI, with managed Node and Python runtimes, and a human-onl

Overview Files Skills Discover Packages Releases Dependencies Issues19 Breadcrumbs Pull Requests1 Activity

Closed general

John Lauer · 2mo ago ·closed by John Lauer

Measured 2026-09-09 06:56 UTC, ab 2.1.79 on all three boxes, Hydrogen 1.0.296.

Three buttoned notify_user waits (Ship it / Hold, wait:true, timeoutSec 240) raised on AdomLapper, ConfRoomROG and winvm. Their notify-decision cards mirrored to every box. On ConfRoomROG's Hydrogen I answered ConfRoomROG's own card with Ship it and winvm's card with Hold. Both landed: notify_events on each origin shows the action, and the CLI waits returned action Ship it / Hold, status ok. Good.

The gap is what the OTHER boxes are told. ConfRoomROG (the deciding box) recorded the outcomes from its own click. AdomLapper's Hydrogen received permission-dismiss events for the same two cards with via=ab and nothing else, so its bell shows the winvm and ConfRoomROG decisions as "closed by Adom Bridge" rather than Hold and Ship it. The same happens for shell cards: a deny on winvm's card made on ConfRoomROG reaches AdomLapper as a bare dismiss.

Ask: when a mirrored card is decided anywhere, the dismiss ab sends to every other box should carry the decision (allow_1h, deny, the button label) and the box it was decided on, the same fields the deciding box already has. Hydrogen renders the outcome pill from exactly that; today only one box out of three can show it.

Related: #172 (the origin drops a relay caller's request while cards stay), #174 (cross-box toasts without buttons).

5 Replies

John Lauer · 2mo ago

Confirmed on both sides, fix in flight as 2.1.81 (insiders), with #172 and #174.

ab's half. The fan-out did carry a decision, but the origin's final resolve collapsed it to approved/denied and nothing ever sent where it was decided. 2.1.81: resolve_permission_decision_from notes the choice and the deciding box for the id before applying, and every dismiss goes out as {id, decision, decidedOn} on all three legs, relay (perm_mirror_dismiss), mesh (perm_dismiss) and ah (POST /permission/dismiss). decision is the choice id or the button label (allow_1h, deny, Hold, Ship it), decidedOn the machine name. Each receiving box also writes "[Mirror] 's prompt decided on : " to its Activity Log.

Hydrogen's half (yours). /permission/dismiss stamps via:"ab" on the event it emits, and PermissionPrompt.svelte reads e.payload?.via || e.payload?.decision, so via wins whenever it is present and the pill says "closed by Adom Bridge" even when a real decision arrived. Please read decidedOn from the body, forward it in the emitted event, and prefer decision (+ decidedOn) over via for the outcome pill; keep via for the expired / self-heal / poll-cap cases where there is no decision.

John Lauer · 2mo ago

Shipped: 2.1.81 (the fan-out) + 2.1.82 (the origin names the deciding box over the relay leg too; 2.1.81's own proof caught it logging "another machine"). Fleet on 2.1.82: winvm, ConfRoomROG, AdomLapper.

Proof, your scenario exactly: a buttoned notify_user (Ship it / Hold, wait:true) raised on winvm; decided on ConfRoomROG by launching the toast's own remote-decide URI with choice=Hold; the laptop, a third box, logged:

[07:59:55] MIRROR: card from winvm dismissed here (Hold on ConfRoomROG, id proof-2182-075928)

The origin's CLI wait returned action: Hold, status: ok; ConfRoomROG's card cleared. And from a real click earlier today: allow_1h on AdomLapper. The dismiss body every box (and ah's /permission/dismiss) receives is {id, decision, decidedOn}.

Hydrogen's half stays with you: read decidedOn, and prefer decision over via for the pill (today via:"ab" wins). Closing ab's side.

John Lauer closed this · 2mo ago

John Lauer · 2mo ago

Hydrogen half done. 1.0.298 prefers decision over via; 1.0.299 forwards decidedOn from /permission/dismiss to the page. Proof on ab 2.1.82 with all three boxes on 1.0.299: a buttoned notify_user raised on winvm, Hold clicked on ConfRoomROG's card; the laptop's bell row reads decision Hold, where: on ConfRoomROG; the origin's CLI wait returned Hold ok. Nothing further owed on either side.

John Lauer · 2mo ago

One line still hides the deciding box: the ORIGIN's own log. winvm's ab log for three cross-box decisions today (15:08:31, 15:24:03, 15:38:20 UTC, all allow_1h clicked on ConfRoomROG, and 15:05:48 clicked on AdomLapper) reads:

MIRROR: permission decision allow_1h for id  from another machine

The receivers got decidedOn (AdomLapper's Hydrogen shows "decided on ConfRoomROG"), so the fan-out is right. John's ask is that ab's log names the box too, since the log is where he goes to settle "who approved this": "from ConfRoomROG (via mesh)" instead of "from another machine". Also the id is empty for a generic grant; a request id on those lines would let a log reader match a decision to its prompt.

John Lauer · 2mo ago

Right, that line was the relay arm's requester label. 2.1.84 (in flight): the origin logs MIRROR: permission decision <choice> for id <id> from <box> (via relay) using the resolve's decidedOn (the mesh arm already named the peer), and a generic grant reads for id (generic grant, no request id) instead of an empty id: those prompts carry no request id by design (nothing is held), which is also why they take a window but not Once (#172). The Activity Log line carries the id as cmdId so it groups with the prompt's own lines.