fix(dashboard): stream AI thinking on every Planning Mode generation step

The workspace loader that covers all follow-up turns (next question,
refine, contextual comments, question regeneration) showed only a spinner
and elapsed time — streamed thinking/output was visible only on the first
pre-summary turn. Reuse the initial loading view's thinking container and
toggle there, and mirror the generation-activity label instead of a
hardcoded "Generating plan…".

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

.changeset/planning-thinking-all-steps.md

@@ -0,0 +1,7 @@
+---
+"@runfusion/fusion": patch
+---
+
+summary: Every Planning Mode generation step now streams AI thinking/output, not just the first turn.
+category: fix
+dev: The planning workspace loader (follow-up turns — next question, refine, contextual comments, question regeneration) reuses the initial loading view's thinking container/toggle and mirrors the generation-activity label.

packages/dashboard/app/components/PlanningModeModal.tsx

@@ -3975,12 +3975,38 @@ export function PlanningModeModal({ isOpen, onClose, onTaskCreated, onTasksCreat
               {view.type === "loading" && (
                 <div className="planning-workspace-loader" role="status" aria-live="polite">
                   <Loader2 size={40} className="spin" />
-                  <strong>{t("planning.generatingPlan", "Generating plan…")}</strong>
+                  <strong>
+                    {generationActivity === "question"
+                      ? t("planning.generatingQuestion", "Generating next question…")
+                      : t("planning.generatingPlan", "Generating plan…")}
+                  </strong>
                   {generationStartTime && <span>{t("planning.thinkingElapsed", "Thinking… ({{seconds}}s)", { seconds: elapsedSeconds })}</span>}
                   <button className="btn planning-stop-btn" type="button" onClick={() => void handleStopGeneration()}>
                     <StopCircle size={14} />
                     <span className="icon-ml-6">{t("planning.stop", "Stop")}</span>
                   </button>
+                  {/*
+                  FNXC:PlanningThinkingVisibility 2026-07-23-22:45:
+                  Every Planning Mode generation step must stream the model's thinking/output to
+                  the operator, not only the first (pre-summary) turn. This workspace loader
+                  covers all follow-up turns — next question, refine, contextual comments, and
+                  question regeneration — and previously showed only a spinner with elapsed
+                  time. Reuse the same thinking container + toggle as the initial loading view.
+                  */}
+                  <div className="planning-thinking-container">
+                    <button
+                      className="planning-thinking-toggle"
+                      onClick={() => setShowThinking(!showThinking)}
+                      type="button"
+                    >
+                      {showThinking ? t("planning.hideThinking", "Hide thinking") : t("planning.showThinking", "Show thinking")}
+                    </button>
+                    {showThinking && streamingOutput && (
+                      <div className="planning-thinking-output" ref={thinkingOutputRef}>
+                        <pre>{streamingOutput}</pre>
+                      </div>
+                    )}
+                  </div>
                 </div>
               )}
             </div>

packages/dashboard/app/components/__tests__/PlanningModeModal.planning-flow.test.tsx

@@ -842,6 +842,30 @@ describe("PlanningModeModal sequential flow", () => {
     expect(await screen.findByText("Thinking… (7s)")).toBeInTheDocument();
     dateNow.mockRestore();
   });
+  /*
+  FNXC:PlanningThinkingVisibility 2026-07-23-22:45:
+  Every generation step must stream thinking/output to the operator. Follow-up turns render
+  the workspace loader (summary present), which previously showed only a spinner + elapsed
+  time; this pins the streamed thinking pane there too.
+  */
+  it("streams thinking in the workspace loader during follow-up generations", async () => {
+    mockFetchAiSession.mockResolvedValue({
+      ...base,
+      status: "generating",
+      currentQuestion: null,
+      result: JSON.stringify(summaryWithRefinements),
+      inputPayload: JSON.stringify({ generationPurpose: "plan_update", generationStartedAt: new Date().toISOString() }),
+    });
+    renderSession();
+
+    await waitFor(() => expect(mockConnectPlanningStream).toHaveBeenCalledWith("session-1", "project-1", expect.any(Object)));
+    const handlers = mockConnectPlanningStream.mock.calls[0]?.[2];
+    act(() => handlers?.onThinking?.("Weighing the tradeoffs between approaches…"));
+
+    expect(await screen.findByText("Weighing the tradeoffs between approaches…")).toBeInTheDocument();
+    fireEvent.click(screen.getByRole("button", { name: "Hide thinking" }));
+    expect(screen.queryByText("Weighing the tradeoffs between approaches…")).toBeNull();
+  });
   it("returns to the prior question without an error when generation is stopped", async () => {
     const priorQuestion = { id: "q-prior", type: "text", question: "What should change?" };
     mockFetchAiSession.mockResolvedValue({