iOS history conflates sandbox liveness with outcome and labels failed sessions Done

[yxlyx]

Summary

The iOS history list can label a failed cube session Done (green check) solely because its sandbox is not currently started. Sandbox liveness is not a reliable task outcome, and a failed sandbox-list request is also treated as if every cube were not started.

Physical-device observation

Environment:

• Graff iOS 0.1
• iPhone 16 Pro Max, iOS 26.5.2
• Account-synced cube session

The physical-device accessibility snapshot showed the session row as:

canvaspilot, Done, codegraff, cube, 3 hr ago

At the same time, the current synced transcript ended in four entries with:

Transport error: The network connection was lost.

The associated sandbox was stopped. A stopped sandbox caused the row to present the work as successfully Done, despite the persisted failed turns.

Root cause

SessionsListView.loadHistory() reconstructs account-history status from sandbox liveness only:

let live = (try? await Gateway.sandboxes()) ?? []
let started = Set(live.filter { $0.state == "started" }.map(\.id))
// ...
s.status = started.contains(sb) ? .idle : .done

This has two incorrect implications:

1. Every stopped, gone, pending, or otherwise non-started sandbox means the agent completed successfully.
2. Any Gateway.sandboxes() error is suppressed to [], which labels every cube-backed row Done.

AppSessionRow has no persisted outcome field, so the list cannot distinguish success, transport failure, interruption, or unknown state.

Expected behavior

• Done must represent a persisted successful terminal outcome, not the absence of a running sandbox.
• A transport-failed or interrupted turn should be surfaced as failed/needs attention.
• Sandbox liveness should be displayed separately from task outcome.
• Failure to load sandbox state should remain unknown and must not become Done.

Acceptance criteria

• A stopped sandbox alone never changes a session outcome to Done.
• A failed sandbox-list request does not manufacture successful statuses.
• Successful, failed/interrupted, active, and unknown outcomes have distinct persisted representations.
• The physical-device case above renders failed/needs attention, not Done.
• Tests cover stopped/gone sandboxes and a failed Gateway.sandboxes() request.

Related

• Cube preview buffers NDJSON and long iOS turns fail with networkConnectionLost #283 tracks the preview buffering that triggered the observed transport failures.
• iOS: decisions-first list - attention-sort + Needs-you section + act from the row #58 and iOS: survive backgrounding - scenePhase reconnect + replay + push #60 cover broader iOS attention sorting and durable reconnect/replay behavior; this issue is the narrower status-integrity invariant.