Omniscio

Interrupted sessions section (sidebar)

A dedicated sidebar section, above Live Sessions, holding the sessions whose process stopped on its own rather than by your hand — a crash, a finish, or an unexpected exit. The header counts them and each row carries a red dot, because Interrupted means something ended that you did not end.

On this page

What it is

When a session's CLI process stops (status flips to ended), the row moves into a dedicated Interrupted section in the project sidebar — rendered above Live Sessions (Paused Sessions sits below Live). Needs You and Pinned still sit above all of these. Since 2026-10-22 a fourth section, Routine updates, sits between Live and Paused (see routine-updates-section.md). The section header shows an ember-colored warning triangle (AlertTriangle icon) and the label "Interrupted (N)" where N is the count. Each session's dot shows red (a true failure/broken state since the 2026-08-13 color consolidation) instead of the default gray, signaling that something stopped unexpectedly and the user may need to take action.

This is different from Paused, which the user chose deliberately. Interrupted means the process died or finished on its own — the app didn't restart, the user didn't click Pause. The section tells the user "these need your attention — resume or archive them."

Errored / stalled sessions now group here too (2026-08-11). A session that hits an error or stalls is no longer treated as an "attention" item — it does not pop into the Inbox, add a project count badge, or play a chime. Instead it collects quietly in this Interrupted section (sorted by recency), exactly like an ended session, so you get to it when you're ready. No exceptions (2026-10-24): a session the app is quietly reconnecting (a dropped CLI) belongs here too, the moment it fails — it keeps its ember "Reconnecting…" dot, and it moves back to Live by itself once it is running again. Before this, a failed session sat under Live for up to five minutes, where it read as active work.

Every failure / give-up / interruption groups here now (2026-08-21). The Interrupted pile is the home for anything that stopped without delivering a final message — a give-up ("Recovery failed", incl. out-of-balance), an API or auth error, a stop you triggered, a dropped/aborted turn, a sub-agent timeout, or an app-close interruption. Needs You is now reserved for sessions that succeeded and need your response (a finished answer, a question, a plan to approve, a permission, a mission, or a merge-conflict handback). So failures no longer pop into the Inbox, chime, or add a badge — they wait quietly here (still a red dot, still one-click retriable). This reverses the 2026-08-14 rule that surfaced "Recovery failed" in Needs You.

A wait that timed out is NOT interrupted — it stays in Live (2026-10-25). When a session ends its turn waiting on something (a background job, a cloud test, a PR, another agent's reply) and the app eventually stops watching that wait, the row reads "wait timed out" with its orange dot and Keep waiting button. That session was never cut off, so it stays under Live, not here, and agents reading GET /sessions see bucket: live, state: waiting instead of crashed. It still never pops into the Inbox or chimes. Owner report: a dozen of these piled up here every time she came back to her computer, so a real interruption could not be told from a session that was just waiting.

An Overseer that handed its post to a successor is NOT interrupted (2026-11-19). When an Overseer hands its post to a new session (its daily restart, a fresh start, or a hand-off it asked for), the old session ends on purpose and a newer one carries on. The app records that hand-over on the old session, so it never shows here: it waits quietly in Routine updates (when "show ended sessions" is on) until the app files it away, and the red interrupted count leaves it out. Agents reading GET /sessions see it as bucket: archived, state: done. This applies to Overseer sessions only: an ordinary session whose hand-off to a new session failed can carry the same marker with no successor behind it, so it still shows here when it stops. An Overseer that stopped with nobody taking over its post still shows here. Owner report: the Interrupted list held five Overseer sessions every day, each one already replaced by a live successor, so real interruptions were buried.

A session the app could not give a working folder is NOT interrupted (2026-11-03). If a session resumes (for example after the app restarts) and the app fails to make it a working folder, it now waits in Needs You with a note saying it was not started, instead of landing here for good. Nothing went wrong with the session itself, and your next message tries again. Live sessions (running, waiting on you, or waiting on a background job) also keep their working folder now, however long they sit idle, so this happens far less often. A session held back to protect its work (a folder with unsaved files it cannot safely reuse) still shows here.

Restarting comes back into the session's own folder (2026-11-04). Once a session's work has landed, the app's pool of spare working folders may take its folder back and then hold it, because the session itself is still sitting in it. Restarting such a session used to fail: the app asked for a second folder on the same branch, git refused because the original folder still had that branch checked out, and the session went straight back to Interrupted with a note that it could not be placed on its own branch — every time you tried. A resumed session whose folder is still checked out on its own branch now simply starts in that folder again, so it comes back instead of failing. A folder that genuinely belongs to somebody else is still never handed over.

The hub row now shows an interrupted count (2026-08-22). Every project/hub row in the sidebar surfaces a red interrupted count — the number of sessions in this same Interrupted pile, drawn from the exact same isInterruptedSectionMember rule, EXCLUDING rate-limit waiting (which keeps its own orange count) and, since 2026-09-24, any pinned or saved session (those sit in their own Pinned / Saved sections, not in this one) — so you can see at a glance which hubs have interrupted sessions without opening each one. Opening a hub loads its whole Interrupted pile (not just its most recent sessions), so the red number and this section's header agree even on a very busy hub. It is a passive glance badge only: no Inbox row, no chime, no taskbar Needs You badge (those stay quiet). See projects-sidebar.md "Status counts on project rows".

Rate-limit-parked (waiting) sessions group here too (2026-08-14). A session paused on a usage limit — the ember "waiting" state that auto-resumes on its own — now collects in this Interrupted section instead of showing under Live (where it read as active work). It stays calm ember and needs nothing from you — though its banner now carries a one-click Stop (2026-08-16) if you'd rather halt it than let it auto-resume (that Stop pauses it, so it won't auto-resume until you Unpause; see pause-or-stop-a-session.md); only its section changed. (It's still never an amber "attention" item and never chimes — the account-capacity rules are unchanged.)

Where to find it

In the project sidebar, in the section headed Interrupted (N) with an ember-coloured warning triangle — rendered above Live Sessions, while Paused Sessions sits below it. Needs You and Pinned still sit above all three.

How it behaves

The Live Sessions header counts the rows the section lists

The Live Sessions header reads "Live Sessions (N)", where N is the number of conversations in the section — or "Live Sessions (N · +S)" while some of them are still spinning up, S being that not-yet-streaming remainder. A conversation that is simply idle between turns counts: a cloud session waiting on its next instruction, and a conversation a teammate shared with you, which sits idle almost all the time because it is the other person's assistant that is between turns. (Owner report 2026-10-31, on a guest's screen: a listed shared chat sat under "Live Sessions (0)".) Three kinds of row are positively not live and are left out — an untouched pre-warmed blank from the instant-new-session pool ("Session Ready"), a session that is tearing down, and a reconnecting error/stalled row, which files under Interrupted anyway. Rate-limit-parked waiting sessions are not listed here at all (since 2026-08-14 they sit in the Interrupted section). It is a display refinement only: the session list, sort order, and keyboard/swipe navigation are unchanged.

The section only appears when the "show ended sessions" setting is on (Settings → Workflow). When off, ended sessions are hidden entirely and the Interrupted header vanishes. The section header is collapsible (click to hide/show its rows), default expanded. Collapse state persists per-project.

How to use it

  • Waiting for a working folder is not Interrupted: a session that is only waiting for a folder of its own stays in Live, shows how many minutes it has waited, and starts by itself as soon as one is free.
  • Opens at the very bottom: clicking an interrupted session always jumps you to the bottom — its latest output and the reason it stopped — and keeps you pinned there while the conversation finishes rendering, so you never land stranded in the middle. (It stays put the instant you scroll up to read back.)
  • Revive a session: click it, then send a message (or click the "Please continue" nudge button). Omniscio auto-resumes the CLI. After reviving, Omniscio auto-advances you to the next interrupted session — see gray-session-triage.md.
  • Archive it: middle-click, right-click → Archive, or press E / Ctrl+W. Per-section advance keeps you in the Interrupted pile (never jumps to Paused).
  • Collapse the section: click the "Interrupted (N)" header to hide its rows. The collapse state is saved per-project.

Visual design

Element Value
Icon AlertTriangle (lucide-react), ember-colored
Header text color text-ember-500 (hover: text-ember-600)
Dot color (STATUS_DISPLAY for ended) bg-status-error / text-status-error
Tooltip on the dot "Session was interrupted — resume or archive"
Status label "Interrupted"

How it differs from Paused

Interrupted Paused
Trigger Process stopped on its own (ended) User deliberately paused
User action needed Yes — resume or archive Optional — unpause when ready
Dot color Red (failure) Blue
Icon AlertTriangle (⚠) Pause

History

Originally added 2026-06-26 as "Suspended" with a gray Moon icon. Renamed to "Interrupted" with an ember AlertTriangle icon on 2026-07-06 after a human reported confusion: "suspended" and "paused" are synonyms, making the two sections appear redundant. The rename clarifies that Interrupted is a warning state (process stopped unexpectedly) while Paused is an intentional hold. Moved above Live (from between Live and Paused) on 2026-07-19 at the user's request, so interrupted sessions surface at the top of the session list instead of below the running ones. The Live header count was refined on 2026-08-05 to also exclude teal ready (not-yet-launched / unsent) blanks — opening a blank you haven't submitted no longer bumps the count, though the blank still shows as a row. On 2026-08-14 (session-bucket redesign Phase 2), rate-limit-parked waiting sessions moved from Live into this Interrupted section (they auto-resume, so they aren't "running") — and the Manage-sessions Stop list gained the same pile grouping so its count reconciles with the sidebar. On 2026-08-23 the per-section auto-advance (act on a row → jump to the next in the same section) was broadened from ended-only to the WHOLE Interrupted section — so acting on a waiting / errored / interrupted-needs_you row now advances to the next Interrupted session and bounces back within the section, matching Needs You and Paused (see gray-session-triage.md). On 2026-10-22 the reconnecting exception was bounded: a reconnecting error/stalled session stayed in Live for at most 5 minutes (RECONNECTING_LIVE_MAX_MS, measured from statusChangedAt), then filed here with its ember dot unchanged. On 2026-10-24 that exception was removed entirely (owner rule: a failed session must never sit under Live) — every error/stalled session files here immediately, its ember "Reconnecting…" dot untouched, and returns to Live on its own once it is running again. The same 2026-10-22 day a new Routine updates section was added below Live. On 2026-10-31 the Live header went back to counting the rows the section lists, instead of counting only the ones actually streaming: a conversation idle between turns — a cloud row, or a guest's shared conversation, which stays ready while the other person's assistant is between turns — sat under "LIVE SESSIONS (0)" (owner report from a live-test guest's screen). The header keeps its +N disclosure for the rows still spinning up, and still leaves out the untouched pre-warmed blank, a teardown, and a reconnecting error/stalled row.

Last verified 2026-11-19