Function askSideQuestion

src/node/services/sideQuestionService.ts:166–511

(
  opts: AskSideQuestionOptions
)

}

export async function askSideQuestion(
  opts: AskSideQuestionOptions
): Promise<Result<AskSideQuestionSuccess, NameGenerationError>> {
  const { workspaceId, question, candidates, aiService, historyService, emitChatEvent } = opts;

  const trimmedQuestion = question.trim();
  if (trimmedQuestion.length === 0) {
    return Err({ type: "unknown", raw: "Side question is empty" });
  }

  if (candidates.length === 0) {
    return Err({ type: "unknown", raw: "No model candidates available for side question" });
  }

  // ---------------------------------------------------------------------
  // 0. Snapshot any in-flight MAIN-AGENT stream so the renderer can later
  //    split the interrupted message into pre-aside + post-aside halves
  //    around this /btw pair.
  //
  // Two structural guarantees make this safe to read synchronously:
  //  - `aiService.getStreamInfo` is a sync getter over an in-memory map
  //    (StreamManager.workspaceStreams), so no race with disk I/O.
  //  - The /btw pipeline bypasses StreamManager entirely (no
  //    `streamManager.startStream` call), so `getStreamInfo` can ONLY
  //    return a main-agent stream — never a concurrent /btw. The
  //    "side question must not interrupt itself" filter is therefore
  //    structural rather than a runtime check.
  //
  // MUST run before the first `await` below: any awaited work between
  // this read and the user-message append widens the racy window where
  // the main agent could finish streaming (StreamingMessageAggregator
  // would then no longer have the interruption anchor we promise).
  // ---------------------------------------------------------------------
  const liveStreamSnapshot =
    opts.liveStreamSnapshot ?? snapshotSideQuestionLiveStream(aiService.getStreamInfo(workspaceId));
  const interruption = liveStreamSnapshot
    ? {
        interruptedMessageId: liveStreamSnapshot.messageId,
        // Text length anchors the split across text parts; part index keeps
        // non-text parts (reasoning/tool/file) that were already visible at
        // the same text offset on the pre-aside side after reload.
        interruptedTextLength: liveStreamSnapshot.parts.reduce(
          (sum, p) => (p.type === "text" ? sum + p.text.length : sum),
          0
        ),
        interruptedPartIndex: liveStreamSnapshot.parts.length,
        interruptedHistorySequence: liveStreamSnapshot.historySequence,
      }
    : undefined;

  // ---------------------------------------------------------------------
  // 1. Persist + emit the user's /btw message FIRST so the question shows
  //    up in the chat immediately. Even if the model call fails, the user
  //    can see what they asked.
  // ---------------------------------------------------------------------
  const rawCommand = `${SIDE_QUESTION_COMMAND} ${trimmedQuestion}`;
  const userMessage: MuxMessage = createMuxMessage(createUserMessageId(), "user", trimmedQuestion, {
    timestamp: Date.now(),