diff --git a/.changeset/fix-interactive-tool-continue.md b/.changeset/fix-interactive-tool-continue.md
new file mode 100644
index 00000000..4e895ea5
--- /dev/null
+++ b/.changeset/fix-interactive-tool-continue.md
@@ -0,0 +1,15 @@
+---
+"perstack": patch
+---
+
+fix(perstack): handle stoppedByInteractiveTool in TUI execution loop
+
+When an expert calls an interactive tool and the run stops with status
+"stoppedByInteractiveTool", the TUI execution loop now correctly:
+
+1. Allows the user to continue by entering input (previously the loop exited)
+2. Treats the user's input as an interactive tool result (previously it was
+   treated as a regular text query)
+
+This fixes the issue where the TUI would hang or display incorrectly when
+an interactive tool was called during a conversation.
diff --git a/apps/perstack/src/start.ts b/apps/perstack/src/start.ts
index 52ea7190..9060b44c 100644
--- a/apps/perstack/src/start.ts
+++ b/apps/perstack/src/start.ts
@@ -157,6 +157,8 @@ export const startCommand = new Command()
       // Phase 3: Execution loop
       let currentQuery: string | null = selection.query
       let currentJobId = currentCheckpoint?.jobId ?? input.options.jobId
+      // Track if the next query should be treated as an interactive tool result
+      let isNextQueryInteractiveToolResult = input.options.interactiveToolCallResult ?? false
 
       // Track accumulated events across continues
       // On first iteration, load all events for the job up to the checkpoint
@@ -192,7 +194,7 @@ export const startCommand = new Command()
             jobId: currentJobId,
             expertKey: selection.expertKey,
             input:
-              input.options.interactiveToolCallResult && currentCheckpoint
+              isNextQueryInteractiveToolResult && currentCheckpoint
                 ? parseInteractiveToolCallResult(currentQuery, currentCheckpoint)
                 : { text: currentQuery },
             experts,
@@ -220,16 +222,20 @@ export const startCommand = new Command()
         const result = await executionResult
 
         // Check if user wants to continue
-        if (
-          result.nextQuery &&
-          (runResult.status === "completed" ||
-            runResult.status === "stoppedByExceededMaxSteps" ||
-            runResult.status === "stoppedByError")
-        ) {
+        const canContinue =
+          runResult.status === "completed" ||
+          runResult.status === "stoppedByExceededMaxSteps" ||
+          runResult.status === "stoppedByError" ||
+          runResult.status === "stoppedByInteractiveTool"
+
+        if (result.nextQuery && canContinue) {
           currentQuery = result.nextQuery
           currentCheckpoint = runResult
           currentJobId = runResult.jobId
 
+          // If the run stopped for interactive tool, the next query is an interactive tool result
+          isNextQueryInteractiveToolResult = runResult.status === "stoppedByInteractiveTool"
+
           // Accumulate events from the completed run for the next iteration
           const newRunEvents = getEventContents(runResult.jobId, runResult.runId)
           if (accumulatedEvents) {