PR Summary

Medium Risk
Moderate risk: refactors client SSE stream processing/reconnect flow and message queue behavior, which can affect chat delivery/order and stop/retry semantics during network drops.

Overview
Improves chat resilience to network drops by introducing a more robust SSE stream reattach/replay flow that tracks eventIds, detects terminal stream statuses, and can reopen the live tail after non-terminal disconnects.

Updates useChat to handle active-stream conflicts by restoring prior optimistic state, rehydrating chat history, and queueing a pending recovery message that is auto-sent once the stream is recovered; UI (MothershipChat, Home, workflow panel) now treats isReconnecting as stream-active for autoscroll/streaming indicators.

Adds extra server-side SSE logging around client disconnect/cancel in chat-streaming.ts, and exports fetchChatHistory for recovery flows.

^{Written by Cursor Bugbot for commit 03d6cab. Configure here.}

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Mar 26, 2026 0:25am

Greptile Summary

This PR adds client-side disconnect resilience for SSE chat streams. When a live stream is interrupted by a network drop, the client now re-attaches by replaying buffered events from a batch endpoint and then opening a fresh SSE tail — looping until the server signals a terminal status. It also handles the race condition where a new message is sent while the server still has an active stream for the same chat (conflict error), rolling back the optimistic UI update and queuing the user's message as a pendingRecoveryMessage to be retried automatically once the in-flight stream finishes.

Key changes:

• use-chat.ts: Extracts attachToExistingStream, fetchStreamBatch, and applyChatHistorySnapshot callbacks; upgrades processSSEStream to track eventId and support replaying into existing state (preserveExistingState); adds pendingRecoveryMessage state for conflict-retry queuing.
• mothership-chat.tsx / home.tsx / panel.tsx: Propagates new isReconnecting prop so the UI correctly disables input and shows streaming indicators during reconnection.
• tasks.ts: Exports fetchChatHistory for direct use during recovery.
• chat-streaming.ts: Extracts markClientDisconnected helper and adds structured logging on stream close/cancel.

One concrete bug found: stopGeneration clears pendingRecoveryMessageRef.current (the ref) but never calls setPendingRecoveryMessage(null) (the React state). This leaves the recovery message permanently stuck in visibleMessageQueue — the "Send Now" and remove buttons both operate on the already-nulled ref and silently do nothing, making the item impossible to dismiss without a page reload.

Confidence Score: 3/5

• Not yet safe to merge — the stopGeneration / pendingRecoveryMessage state mismatch will permanently wedge the queue UI in specific but reachable user flows.
• The architectural approach is sound and the happy-path reconnect logic is well-structured. Five of the six changed files are clean. The single concrete P1 bug — stopGeneration clearing only the ref and not the React state for pendingRecoveryMessage — produces a permanently stuck UI element that cannot be dismissed, which is a visible regression in the conflict-recovery path. A one-line fix (setPendingRecoveryMessage(null)) resolves it.
• apps/sim/app/workspace/[workspaceId]/home/hooks/use-chat.ts — specifically the stopGeneration function and the two navigation reset effects.

Important Files Changed

Sequence Diagram

sequenceDiagram
    participant U as User
    participant SC as sendMessage
    participant ATES as attachToExistingStream
    participant FSB as fetchStreamBatch
    participant PSS as processSSEStream
    participant SG as stopGeneration
    participant F as finalize

    U->>SC: send message
    SC->>SC: POST /api/copilot/chat
    alt Primary stream ends cleanly
        SC->>PSS: processSSEStream(reader, assistantId)
        PSS-->>SC: { sawStreamError, sawDoneEvent, lastEventId }
        SC->>FSB: fetchStreamBatch(streamId, lastEventId, signal)
        FSB-->>SC: { status, events }
        alt Terminal status
            SC->>F: finalize()
        else Non-terminal (network drop)
            SC->>ATES: attachToExistingStream(...)
            loop Until terminal status or abort
                ATES->>PSS: processSSEStream(replayStream)
                PSS-->>ATES: termination result
                ATES->>ATES: open live SSE tail
                PSS-->>ATES: live termination result
                ATES->>FSB: fetchStreamBatch(next batch)
                FSB-->>ATES: { status, events }
            end
            ATES-->>SC: { aborted, error }
            SC->>F: finalize()
        end
    else Conflict error (stream already active)
        SC->>SC: rollback optimistic state
        SC->>SC: setPendingRecoveryMessage(queuedMessage)
        SC->>SC: preparePendingStreamRecovery(chatId)
        SC->>ATES: attachToExistingStream(existing stream)
        ATES-->>SC: { aborted, error }
        SC->>F: finalize()
        F->>F: detect pendingRecoveryMessage
        F->>SC: sendMessage(recoveryMessage) [via queueMicrotask]
    end

    U-->>SG: stop generation (optional)
    SG->>SG: abort controller, clear refs
    Note over SG: ⚠️ pendingRecoveryMessage state NOT cleared

_{Reviews (2): Last reviewed commit: "address queued message conflicts during ..." | Re-trigger Greptile}

bugbot run

bugbot run

@greptile

bugbot run

bugbot run

bugbot run

bugbot run