diff --git a/app/lib/workflows/runAgentStep.ts b/app/lib/workflows/runAgentStep.ts
index aebebfe46..33db67143 100644
--- a/app/lib/workflows/runAgentStep.ts
+++ b/app/lib/workflows/runAgentStep.ts
@@ -36,6 +36,12 @@ export type RunAgentStepInput = {
    * is added to `experimental_context` right before each model call.
    */
   agentContext: DurableAgentContext;
+  /**
+   * Whether a user is present to answer `ask_user_question`. Interactive chat
+   * sets true (default); headless runs (`/api/chat/runs`, `customer-prompt-task`)
+   * set false so the tool is withheld — no one is there to reply.
+   */
+  interactive?: boolean;
   /**
    * Stable id to assign to the assistant message produced by this
    * step. Generated once in `runAgentWorkflow` so:
@@ -107,7 +113,7 @@ export async function runAgentStep(input: RunAgentStepInput): Promise<RunAgentSt
   // without it, a hung tool keeps streamText awaiting forever on stop.
   const tools = wrapToolsWithAbort(
     addCacheControlToTools({
-      tools: buildAgentTools({ skills: input.agentContext.skills }),
+      tools: buildAgentTools({ skills: input.agentContext.skills, interactive: input.interactive }),
       model: input.modelId,
     }),
     cancelController.signal,
diff --git a/app/lib/workflows/runAgentWorkflow.ts b/app/lib/workflows/runAgentWorkflow.ts
index c0cdec7ff..5a4d27170 100644
--- a/app/lib/workflows/runAgentWorkflow.ts
+++ b/app/lib/workflows/runAgentWorkflow.ts
@@ -28,6 +28,8 @@ export type RunAgentWorkflowInput = {
    */
   accountId: string;
   modelId: string;
+  /** Whether an interactive user is present (chat UI) to answer ask_user_question. */
+  interactive?: boolean;
   /**
    * Optional chat title — used as context for the auto-commit
    * message-generation LLM call.
diff --git a/lib/agent/__tests__/buildAgentTools.test.ts b/lib/agent/__tests__/buildAgentTools.test.ts
index e684818f2..4f155c91d 100644
--- a/lib/agent/__tests__/buildAgentTools.test.ts
+++ b/lib/agent/__tests__/buildAgentTools.test.ts
@@ -69,4 +69,16 @@ describe("buildAgentTools", () => {
     const tools = buildAgentTools() as Record<string, { execute?: unknown }>;
     expect(tools.ask_user_question?.execute).toBeUndefined();
   });
+  it("omits `ask_user_question` in non-interactive (headless) mode — no user to answer", () => {
+    const tools = buildAgentTools({ interactive: false });
+    expect(tools).not.toHaveProperty("ask_user_question");
+    for (const name of ALWAYS_PRESENT.filter(n => n !== "ask_user_question")) {
+      expect(tools).toHaveProperty(name);
+    }
+  });
+
+  it("keeps `ask_user_question` when interactive (default and explicit true)", () => {
+    expect(buildAgentTools()).toHaveProperty("ask_user_question");
+    expect(buildAgentTools({ interactive: true })).toHaveProperty("ask_user_question");
+  });
 });
diff --git a/lib/agent/buildAgentTools.ts b/lib/agent/buildAgentTools.ts
index 728334b11..1ca91a2db 100644
--- a/lib/agent/buildAgentTools.ts
+++ b/lib/agent/buildAgentTools.ts
@@ -38,8 +38,13 @@ import type { SkillMetadata } from "@/lib/skills/skillTypes";
  * @param options.skills - Discovered skill catalog. When empty / undefined,
  *   `skill` is omitted from the tool record so the model doesn't see it.
  */
-export function buildAgentTools(options: { skills?: SkillMetadata[] } = {}) {
+export function buildAgentTools(options: { skills?: SkillMetadata[]; interactive?: boolean } = {}) {
   const hasSkills = (options.skills?.length ?? 0) > 0;
+  // ask_user_question has no server execute — only a streaming chat UI can
+  // fulfill it. In headless/async runs (customer-prompt-task, /api/chat/runs)
+  // there is no user to answer, so it's a dead-end; omit it there and force the
+  // agent to act (send an honest result or stop) rather than hang on a question.
+  const interactive = options.interactive ?? true;
   return {
     bash: bashTool,
     read: readFileTool,
@@ -50,7 +55,7 @@ export function buildAgentTools(options: { skills?: SkillMetadata[] } = {}) {
     todo_write: todoWriteTool,
     web_fetch: webFetchTool,
     task: taskTool,
-    ask_user_question: askUserQuestionTool,
+    ...(interactive ? { ask_user_question: askUserQuestionTool } : {}),
     ...(hasSkills ? { skill: skillTool } : {}),
   };
 }
diff --git a/lib/chat/buildRunAgentInput.ts b/lib/chat/buildRunAgentInput.ts
index 0630bf562..cc487ddd9 100644
--- a/lib/chat/buildRunAgentInput.ts
+++ b/lib/chat/buildRunAgentInput.ts
@@ -23,6 +23,8 @@ export type BuildRunAgentInputParams = {
    * `/api/chat/runs`). Omitted when absent so the service key never leaks.
    */
   recoupAccessToken?: string;
+  /** True for interactive chat (default), false for headless runs (withholds ask_user_question). */
+  interactive?: boolean;
   /**
    * Row id of an ephemeral key minted for a headless run, so the workflow can
    * delete it on run end (recoupable/chat#1813). Interactive callers omit it.
@@ -49,6 +51,7 @@ export function buildRunAgentInput({
   skills,
   recoupAccessToken,
   ephemeralKeyId,
+  interactive,
 }: BuildRunAgentInputParams): RunAgentWorkflowInput {
   const repoIds = parseGitHubRepoIdentifiers(cloneUrl);
   const recoupOrgId = cloneUrl ? (extractOrgId(cloneUrl) ?? undefined) : undefined;
@@ -60,6 +63,7 @@ export function buildRunAgentInput({
     accountId,
     modelId,
     sessionTitle,
+    interactive,
     repoOwner: repoIds?.owner,
     repoName: repoIds?.repo,
     agentContext: {
diff --git a/lib/chat/runs/handleStartChatRun.ts b/lib/chat/runs/handleStartChatRun.ts
index 2db6a305a..b9d0125a2 100644
--- a/lib/chat/runs/handleStartChatRun.ts
+++ b/lib/chat/runs/handleStartChatRun.ts
@@ -64,6 +64,7 @@ export async function handleStartChatRun(request: NextRequest): Promise<Response
         skills: provisioned.skills,
         recoupAccessToken: rawKey,
         ephemeralKeyId: keyId,
+        interactive: false,
       }),
     ]);