fix(sidebar): stop reporting an interrupted agent as done (STA-5357)

An interrupted turn rendered on the worktree card and the tab glyph as `done` —
visually identical to a clean completion. A cancelled or dead turn read as
finished work, and with several agents it contributed no signal at all, so a
sibling that merely finished could hide it entirely.

`interrupted` is a flag clamped onto `done` at parse time, never its own state,
so the activity summary never even received it: its Pick was
{state, workingMode}, and the `done` branch swallowed it into hasLiveDone.

Adds the flag end to end — summary, both activity hooks, the resolver, the card
glyph and the tab badge — and slots it below permission and above working, which
is where SUMMARY_STATE_ORDER already ranked it for the text summary. The card and
the expanded agent list now agree.

The tab-bar test that asserted the old behavior is flipped rather than deleted:
it explicitly pinned `done` to mirror the card, and that premise is gone.
This commit is contained in:
Brennan Benson
2026-08-24 15:29:50 -07:00
parent ec4687c434
commit a601e8407a
8 changed files with 139 additions and 8 deletions
@@ -53,6 +53,21 @@ const StatusIndicator = React.memo(function StatusIndicator({
)
}
if (status === 'interrupted') {
return (
<span
className={cn('inline-flex h-3 w-3 shrink-0 items-center justify-center', className)}
title={resolvedTitle}
{...rest}
>
{/* Why red, and why not the emerald done dot: the turn was cancelled or died, so it needs a
human to restart or look — presenting it as a completion is the bug (STA-5357). Matches the
agent row's own interrupted treatment (AgentStateDot uses bg-red-500). */}
<span className="block size-1.5 rounded-full bg-red-500" />
</span>
)
}
if (status === 'permission') {
return (
<span
@@ -26,6 +26,7 @@ export function useWorktreeActivityStatus(worktreeId: string): WorktreeStatus {
hasPermission,
hasLiveWorking,
hasLiveMonitoring,
hasInterrupted,
hasLiveDone,
hasRetainedDone,
agentStatusPaneIdsByTabId
@@ -46,6 +47,7 @@ export function useWorktreeActivityStatus(worktreeId: string): WorktreeStatus {
hasPermission,
hasLiveWorking,
hasLiveMonitoring,
hasInterrupted,
hasLiveDone,
hasRetainedDone
}),
@@ -59,6 +61,7 @@ export function useWorktreeActivityStatus(worktreeId: string): WorktreeStatus {
hasPermission,
hasLiveWorking,
hasLiveMonitoring,
hasInterrupted,
hasLiveDone,
hasRetainedDone
]
@@ -34,6 +34,7 @@ export function selectWorktreeActivityStatuses(
hasPermission,
hasLiveWorking,
hasLiveMonitoring,
hasInterrupted,
hasLiveDone,
hasRetainedDone,
agentStatusPaneIdsByTabId
@@ -50,6 +51,7 @@ export function selectWorktreeActivityStatuses(
hasPermission,
hasLiveWorking,
hasLiveMonitoring,
hasInterrupted,
hasLiveDone,
hasRetainedDone
})
@@ -16,6 +16,9 @@ export type WorktreeAgentActivitySummary = {
hasPermission: boolean
hasLiveWorking: boolean
hasLiveMonitoring: boolean
/** A fresh hook entry carrying `interrupted`. Separate from `hasLiveDone` because the flag only ever
* rides on `done` — folding it in reports a cancelled turn as a clean finish (STA-5357). */
hasInterrupted: boolean
hasLiveDone: boolean
hasRetainedDone: boolean
agentStatusPaneIdsByTabId: Record<string, ReadonlySet<string>>
@@ -27,6 +30,7 @@ const EMPTY_SUMMARY: WorktreeAgentActivitySummary = {
hasPermission: false,
hasLiveWorking: false,
hasLiveMonitoring: false,
hasInterrupted: false,
hasLiveDone: false,
hasRetainedDone: false,
agentStatusPaneIdsByTabId: EMPTY_AGENT_STATUS_PANE_IDS_BY_TAB_ID
@@ -180,6 +184,7 @@ function summariesEqual(
previous.hasPermission === next.hasPermission &&
previous.hasLiveWorking === next.hasLiveWorking &&
previous.hasLiveMonitoring === next.hasLiveMonitoring &&
previous.hasInterrupted === next.hasInterrupted &&
previous.hasLiveDone === next.hasLiveDone &&
previous.hasRetainedDone === next.hasRetainedDone &&
agentStatusPaneIdsByTabIdEqual(
@@ -217,10 +222,14 @@ function agentStatusPaneIdsByTabIdEqual(
function applyLiveAgentState(
summary: WorktreeAgentActivitySummary,
entry: Pick<AgentStatusEntry, 'state' | 'workingMode'>
entry: Pick<AgentStatusEntry, 'state' | 'workingMode' | 'interrupted'>
): void {
if (entry.state === 'blocked' || entry.state === 'waiting') {
summary.hasPermission = true
} else if (entry.interrupted === true) {
// Why before the `done` branch: `interrupted` is clamped to `done` at parse time, so a cancelled
// turn would otherwise set `hasLiveDone` and render as a completed one.
summary.hasInterrupted = true
} else if (entry.state === 'working') {
if (entry.workingMode === 'monitoring') {
summary.hasLiveMonitoring = true
@@ -110,7 +110,11 @@ describe('resolveTerminalTabActivityStatus', () => {
).toBe('done')
})
it('treats an interrupted done as done, matching the worktree card', () => {
// Why this flipped (STA-5357): it used to assert `done`, deliberately mirroring the worktree card,
// which had no interrupted state. That made a cancelled turn indistinguishable from a clean finish
// on every glanceable surface. The card now distinguishes it, so the tab follows rather than
// preserving a consistency that was consistently wrong.
it('reports an interrupted done as interrupted, matching the worktree card', () => {
const interrupted = entry(FIRST_LEAF_ID, 'done', { interrupted: true })
expect(
resolveTerminalTabActivityStatus({
@@ -118,7 +122,22 @@ describe('resolveTerminalTabActivityStatus', () => {
agentStatusByPaneKey: { [interrupted.paneKey]: interrupted },
ptyIdsByTabId: LIVE_PTY
})
).toBe('done')
).toBe('interrupted')
})
it('does not let a finished sibling mask an interrupted pane', () => {
const interrupted = entry(FIRST_LEAF_ID, 'done', { interrupted: true })
const finished = entry(SECOND_LEAF_ID, 'done')
expect(
resolveTerminalTabActivityStatus({
tab: TAB,
agentStatusByPaneKey: {
[interrupted.paneKey]: interrupted,
[finished.paneKey]: finished
},
ptyIdsByTabId: LIVE_PTY
})
).toBe('interrupted')
})
it('falls back to a live working title when hook status is stale', () => {
@@ -20,6 +20,7 @@ type TerminalTabActivityFlags = {
hasPermission: boolean
hasLiveWorking: boolean
hasLiveMonitoring: boolean
hasInterrupted: boolean
hasLiveDone: boolean
paneIds: Set<string>
}
@@ -81,10 +82,12 @@ function getTerminalTabActivityFlags(
} else {
flags.hasLiveWorking = true
}
} else if (entry.interrupted === true) {
// Why before `done` (STA-5357): the flag is clamped to `done`, so folding it in reported a
// cancelled turn as a completed one. This used to mirror the WorktreeCard dot deliberately —
// that dot now distinguishes interrupted, so the tab follows rather than diverging.
flags.hasInterrupted = true
} else if (entry.state === 'done') {
// Why: an interrupted `done` still reads as completed here, matching the
// WorktreeCard dot (resolveWorktreeStatus has no interrupted state); only
// the smart-sort ordering treats interrupts as idle.
flags.hasLiveDone = true
}
}
@@ -103,6 +106,7 @@ function getOrCreateTerminalTabActivityFlags(
hasPermission: false,
hasLiveWorking: false,
hasLiveMonitoring: false,
hasInterrupted: false,
hasLiveDone: false,
paneIds: new Set()
}
@@ -164,6 +168,7 @@ export function resolveTerminalTabActivityStatus({
hasPermission: flags?.hasPermission ?? false,
hasLiveWorking: flags?.hasLiveWorking ?? false,
hasLiveMonitoring: flags?.hasLiveMonitoring ?? false,
hasInterrupted: flags?.hasInterrupted ?? false,
hasLiveDone: flags?.hasLiveDone ?? false,
// Why: retained/orchestration promotions are worktree-aggregate concerns;
// a tab reflects its own live panes and title only.
@@ -180,7 +185,13 @@ export function isTerminalTabActivityLive(status: TerminalTabActivityStatus): bo
* Glyph-bearing attention states for a terminal tab (tab bar + Cmd+J recent chats).
* Quiet active/inactive map to null so identity icons stay clean.
*/
export type TerminalTabAttentionBadge = 'working' | 'monitoring' | 'permission' | 'unread' | 'done'
export type TerminalTabAttentionBadge =
| 'working'
| 'monitoring'
| 'permission'
| 'interrupted'
| 'unread'
| 'done'
/**
* Single priority ladder shared by the tab strip and Cmd+J recent rows:
@@ -202,6 +213,10 @@ export function resolveTerminalTabAttentionBadge({
if (status === 'monitoring') {
return 'monitoring'
}
// Why above unread and done: a cancelled turn needs a human, and an unread bell would bury it.
if (status === 'interrupted') {
return 'interrupted'
}
if (hasUnread) {
return 'unread'
}
@@ -214,11 +229,12 @@ export function resolveTerminalTabAttentionBadge({
/** Map a container activity status onto AgentStateDot's vocabulary (no unread — that's a bell). */
export function terminalTabActivityToAgentDotState(
status: TerminalTabActivityStatus
): 'working' | 'monitoring' | 'permission' | 'done' | null {
): 'working' | 'monitoring' | 'permission' | 'interrupted' | 'done' | null {
switch (status) {
case 'working':
case 'monitoring':
case 'permission':
case 'interrupted':
case 'done':
return status
case 'active':
@@ -0,0 +1,57 @@
import { describe, expect, it } from 'vitest'
import { resolveWorktreeStatus } from './worktree-status'
// Why these live apart from the main suite: they pin the ONE rule STA-5357 adds —
// an interrupted agent is not a completed one, and it must not be masked by a
// sibling that merely finished.
const base = {
tabs: [] as never[],
browserTabs: [] as never[],
ptyIdsByTabId: {},
hasPermission: false,
hasLiveWorking: false,
hasLiveDone: false,
hasRetainedDone: false
}
describe('resolveWorktreeStatus — interrupted (STA-5357)', () => {
it('reports interrupted rather than done for an interrupted agent', () => {
// Why `hasLiveDone` is also set: `interrupted` only ever coexists with `done`
// (agent-status-types.ts clamps it), so the done flag is always present too.
// Before the fix that made the card indistinguishable from a clean finish.
expect(resolveWorktreeStatus({ ...base, hasInterrupted: true, hasLiveDone: true })).toBe(
'interrupted'
)
})
it('keeps interrupted visible when a sibling agent merely finished', () => {
expect(resolveWorktreeStatus({ ...base, hasInterrupted: true, hasLiveDone: true })).not.toBe(
'done'
)
})
it('outranks working, so a cancelled turn is not hidden by a busy sibling', () => {
expect(resolveWorktreeStatus({ ...base, hasInterrupted: true, hasLiveWorking: true })).toBe(
'interrupted'
)
})
it('yields to permission — a prompt waiting on the user is more urgent', () => {
expect(resolveWorktreeStatus({ ...base, hasInterrupted: true, hasPermission: true })).toBe(
'permission'
)
})
it('outranks monitoring', () => {
expect(resolveWorktreeStatus({ ...base, hasInterrupted: true, hasLiveMonitoring: true })).toBe(
'interrupted'
)
})
it('leaves every other combination alone', () => {
expect(resolveWorktreeStatus({ ...base, hasLiveDone: true })).toBe('done')
expect(resolveWorktreeStatus({ ...base, hasLiveWorking: true })).toBe('working')
expect(resolveWorktreeStatus({ ...base, hasLiveMonitoring: true })).toBe('monitoring')
expect(resolveWorktreeStatus({ ...base, hasPermission: true })).toBe('permission')
})
})
+10
View File
@@ -16,6 +16,7 @@ export type WorktreeStatus =
| 'working'
| 'monitoring'
| 'permission'
| 'interrupted'
| 'done'
| 'inactive'
@@ -31,6 +32,7 @@ const STATUS_LABELS: Record<WorktreeStatus, string> = {
working: 'Working',
monitoring: 'Monitoring background tasks',
permission: 'Needs permission',
interrupted: 'Interrupted',
done: 'Done',
inactive: 'Inactive'
}
@@ -143,6 +145,7 @@ export function resolveWorktreeStatus(args: {
hasPermission: boolean
hasLiveWorking: boolean
hasLiveMonitoring?: boolean
hasInterrupted?: boolean
hasLiveDone: boolean
hasRetainedDone: boolean
}): WorktreeStatus {
@@ -164,6 +167,13 @@ export function resolveWorktreeStatus(args: {
if (heuristic === 'permission') {
return 'permission'
}
// Why above working (STA-5357): `interrupted` only ever rides on `done`, so before this it set
// `hasLiveDone` and the card rendered a cancelled turn as a clean finish — and a busy sibling could
// hide it entirely. It sits below permission because a prompt waiting on the user is the more urgent
// ask, and matches SUMMARY_STATE_ORDER so the card and the expanded agent list agree.
if (args.hasInterrupted) {
return 'interrupted'
}
// Why: restored cards get the hook snapshot before panes mount; trust the explicit working row so they stay yellow on restart.
if (args.hasLiveWorking || heuristic === 'working') {
return 'working'