fix(mobile): force refresh active sessions

[iscekic]

Summary

What

• Route session-list manual refresh through the active-sessions live-sync owner when attached.
• Add focused coverage for missed reconnect, cancelled/re-kicked fetches, stuck failed reasons, and no-owner fallback.

Why

After a long background suspension, a missed connection transition can leave the Active now tray stale or empty. A plain query refetch can race with the live-sync owner and be cancelled.

How

• Add a serialized manual refresh entry point and attached-owner registry.
• Make useAgentSessions().refetch() choose the owner path or the plain-query fallback, never both.
• Preserve existing polling, WebSocket handling, and UI states.

Verification

• apps/mobile: pnpm test (297 files, 2583 tests)
• apps/mobile: pnpm format && pnpm typecheck && pnpm lint && pnpm check:unused
• pnpm typecheck
• git diff --check
• iOS E2E: VERIFICATION PASSED. — live Active now tray, pull-to-refresh with a live tray, and pull-to-refresh with an empty tray.
• The long-background wedge is covered by unit tests because iOS simulator E2E cannot deterministically reproduce OS-coalesced reconnect events.

Visual Changes

N/A — no UI change; this changes the manual-refresh code path only.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Executive Summary

No code changed since the previous review at 1889a099: the only new commit (8543ed6, "chore: retrigger review") is empty, so the incremental review scope contains zero changed files and no new findings; the three prior findings were reviewed and rejected by the author with concrete reasoning and are not carried forward.

Files Reviewed (0 files)

• No files changed between 1889a099 and 8543ed6 (git diff is empty)

Previous findings reconciliation

File	Line	Disposition
apps/mobile/src/lib/active-sessions-live-sync.ts	158	Rejected by author: refreshNow() resolving true on a swallowed owner-fetch failure is the plan's accepted AC2 behavior
apps/mobile/src/lib/active-sessions-live-sync.ts	150	Rejected by author: the unbounded queue-drain loop is required so the promise stays pending across write-cancel and replacement fetch
apps/mobile/src/lib/hooks/use-agent-sessions.ts	265	Rejected by author: all current callers pass useOrganization().organizationId, matching the live-sync mount context

These threads remain on the PR for human resolution; they are not re-reported here because they fall outside the incremental changed-line scope.

Assumptions

• Verified against head 8543ed6 and confirmed the retrigger commit introduces no tree changes; no tests were executed (read-only review).

Previous Review Summary (commit 1889a09)

Current summary above is authoritative. Previous snapshots are kept for context only.

Previous review (commit 1889a09)

Status: 3 Issues Found | Recommendation: Address before merge

Executive Summary

ActiveSessionsLiveSync.refreshNow() returns true even when the forced fetch fails, so the plain-refetch fallback in useAgentSessions().refetch() only ever covers the detached case and a failed pull-to-refresh silently leaves the Active-now tray stale with 'manual' stuck in pendingReasons.

Overview

Severity	Count
CRITICAL	0
WARNING	1
SUGGESTION	2

Issue Details (click to expand)

WARNING

File	Line	Issue
apps/mobile/src/lib/active-sessions-live-sync.ts	158	refreshNow() returns true after a failed/externally-cancelled fetch, making the documented active.refetch() fallback unreachable and leaving 'manual' pending, which permanently disables processFetchQueue's empty-reason short-circuit

SUGGESTION

File	Line	Issue
apps/mobile/src/lib/active-sessions-live-sync.ts	150	Follow-the-chain while loop has no hop cap or timeout; write-driven cancels can keep it iterating while the caller's refresh spinner stays up
apps/mobile/src/lib/hooks/use-agent-sessions.ts	265	The "owner is keyed to the same organization context" invariant is a convention, not enforced — no query key is passed or compared

Files Reviewed (3 files)

• apps/mobile/src/lib/active-sessions-live-sync.ts - 2 issues
• apps/mobile/src/lib/hooks/use-agent-sessions.ts - 1 issue
• apps/mobile/src/lib/active-sessions-live-sync.manual-refresh.test.ts - 0 issues

Verified and not flagged

• Module-scoped attachedSync registry: correct across context switches, StrictMode double-invocation, and Fast Refresh. React runs the mount effect's cleanup before the next setup, the mount constructs a fresh instance per run (so attach()'s "already attached" throw cannot fire), and detach()'s attachedSync === this guard prevents a stale owner from nulling a live registration. No reference leak; the dev-only Fast-Refresh module reset degrades to the plain-refetch path.
• Query-key alignment today: all three useAgentSessions callers pass organizationId from the same useOrganization() context the mount reads, and buildActiveSessionsInput collapses undefined to null, so the keys hash identically.
• fetchQueue rejection: every statement in processFetchQueue outside the try/catch is non-throwing in practice (new Set, Date.now, getQueryData + pure predicate, waiter callbacks, kickFetch's void cancelQueries), so the newly-awaited chain does not reject today.
• New test file: afterEach detaches, so the module registry does not leak across tests; Vitest per-file isolation is on. The __hasPendingFetch() assertions do depend on fake-microtask ordering (notifyFetchStart fires before the fake assigns pendingFetch), so they would flip if makeQueryFn became synchronous — worth knowing, not flagged.

Assumptions

• Reviewed at head 1889a099; no tests were executed (read-only review), so the PR's reported pnpm test / typecheck / lint results were taken as stated.
• React Query v5 cancel semantics (cancelRefetch: true default on refetchQueries/invalidateQueries, paused fetches under networkMode: 'online') were reasoned about from the pinned version's documented behavior rather than executed.

Fix these issues in Kilo Cloud

---

_{Reviewed by claude-opus-5 · Input: 20 · Output: 3.1K · Cached: 344.5K}

_{Review guidance: REVIEW.md from base branch main}