[Devbook drift] .devbook/domain/sessions/features.md#session-inventory #878

Scheduled Devbook verify run, 2026-09-30 (devbook:verify-change, kind feature).

Evidence: Sub-feature Sessions from another machine says session records sync behind "a switch of its own ... separately from anything else that syncs"; SessionSyncWorker.ShouldRun gates on SyncFeatures.Sync alone (line 273), the one key that also syncs tasks, and AppFeatures.cs folds session-sync into sync as a FormerKeys entry. Everything else matches: _view = AgentSessionView.Live reset per pane (SessionsPaneTests.Live_is_the_view_the_pane_opens_in, Reopening_the_pane_opens_on_live_again), Live = Running or Stalled (AgentSessionViews.IsLive, A_stalled_session_is_live_because_silence_is_not_an_ending), resolved repository marked and titled (RepositoryTitle, A_recorded_repository_is_shown_over_a_placed_one_and_either_scopes), PRs/tokens/entrypoint (Tokens_are_output_and_input_..., The_entrypoint_is_a_quiet_badge_beside_the_type), recorded rows + Update records (A_recorded_row_says_it_comes_from_the_saved_record_and_keeps_its_folder, Update_records_amends_everything_and_says_what_it_did), no polling (only RefreshAsync re-reads). The code also carries things the chapter does not: see below.

Detail

• The chapter (Sessions from another machine) says session records sync behind "a switch of its own, and it starts off", switched on separately from anything else that syncs.
• The code has one switch. SessionSyncWorker.ShouldRun => _features.IsEnabled(SyncFeatures.Sync) (src/Infrastructure/Backlog.Infrastructure.Sync/Sessions/SessionSyncWorker.cs:273). AppFeatures.cs lists FormerKeys: ["device-pairing", "task-sync", "session-sync"] on sync, so tasks and session records go together.
• Side finding: the user-facing sync description in AppFeatures.cs says a session's title never travels. SessionRecordMapping.cs:93 does send it (Cut(session.Title, ...)), and the chapter agrees with the code. The UI string is wrong.
• Code-ahead parts the chapter lacks: the Machine filter (AgentSessionEnvironments.On; SessionsPaneTests.Choosing_an_environment_keeps_only_its_rows_and_says_how_many_of_how_many), and opening the pane on a session a task names, which widens the filters (FocusSessionId, RevealFocused; A_session_a_task_asks_for_is_shown_marked_and_focused_even_when_live_hides_it, HomeWorkspaceSurfaceTests.Opening_a_session_from_a_task_shows_the_sessions_surface_on_that_session).
• Wording: the chapter says a missing repository reads "not recorded". The pane renders an em dash (Or(), Missing = "—"; What_an_agent_did_not_record_is_an_em_dash).
• Question: should session-record sync be its own switch, as the chapter says, or is the single sync switch (see the SyncFeatures.Sync rationale) the agreed model, so the chapter should change?

Action: a person decides which side is the model: the chapter (then the code is a defect, for apply-change) or the code (then the chapter is stale, for capture-specs). The question is under Detail above.

Read-only: nothing was run or written. The chapter carries no draft or proposed status, and no open pull request or change names it.