From e61efb6cd8a79915041652238be00efbf50b0aa5 Mon Sep 17 00:00:00 2001
From: Peter Steinberger <steipete@gmail.com>
Date: Mon, 28 Sep 2026 22:46:22 -0700
Subject: [PATCH] fix(agents): media generation blocks channel turns sharing
 the main session (#160998)

Pass the durable runSessionKey to image/video/music tools so shared-main channel sessions detach media generation again (regression from #159179), and wait for the terminal image task state in the image-generation-lifecycle e2e.
---
 src/agents/openclaw-tools.ts                  |  6 +----
 src/agents/openclaw-tools.tts-config.test.ts  | 11 +++++----
 .../image-generation-lifecycle.e2e.test.ts    | 23 +++++++++++++++++++
 3 files changed, 31 insertions(+), 9 deletions(-)

diff --git a/src/agents/openclaw-tools.ts b/src/agents/openclaw-tools.ts
index a8c27904da22..f1a221f7906a 100644
--- a/src/agents/openclaw-tools.ts
+++ b/src/agents/openclaw-tools.ts
@@ -146,10 +146,6 @@ export function createOpenClawTools(options?: OpenClawToolsOptions): AnyAgentToo
   });
   const trimmedRunSessionKey = options?.runSessionKey?.trim();
   const requesterSessionKey = trimmedRunSessionKey || options?.agentSessionKey;
-  const mediaGenerationAgentSessionKey =
-    trimmedRunSessionKey && isCronRunSessionKey(trimmedRunSessionKey)
-      ? trimmedRunSessionKey
-      : options?.agentSessionKey;
   const imageTool =
     options?.agentDir &&
     resolveImageToolFactoryAvailable({
@@ -182,7 +178,7 @@ export function createOpenClawTools(options?: OpenClawToolsOptions): AnyAgentToo
     config: options?.config,
     agentDir: options?.agentDir,
     authProfileStore: options?.authProfileStore,
-    agentSessionKey: mediaGenerationAgentSessionKey,
+    agentSessionKey: requesterSessionKey,
     requesterAgentId: sessionAgentId,
     requesterOrigin: widgetPresentation.deliveryContext ?? undefined,
     workspaceDir,
diff --git a/src/agents/openclaw-tools.tts-config.test.ts b/src/agents/openclaw-tools.tts-config.test.ts
index 5f50c0c7069f..2adde6170cb6 100644
--- a/src/agents/openclaw-tools.tts-config.test.ts
+++ b/src/agents/openclaw-tools.tts-config.test.ts
@@ -180,16 +180,19 @@ describe("createOpenClawTools context wiring", () => {
     expect(onYield).not.toHaveBeenCalled();
   });
 
-  it("uses the isolated cron run key for background media completions", () => {
+  it.each([
+    ["agent:main:cron:daily-media", "agent:main:cron:daily-media:run:run-123"],
+    ["agent:main:qa-channel:default:direct:media-requester", "agent:main:main"],
+  ])("uses the durable run key for background media from %s", (agentSessionKey, runSessionKey) => {
     createTools({
       config: mediaConfig,
-      agentSessionKey: "agent:main:cron:daily-media",
-      runSessionKey: "agent:main:cron:daily-media:run:run-123",
+      agentSessionKey,
+      runSessionKey,
       onYield: vi.fn(),
     });
     for (const factory of [mocks.image, mocks.video, mocks.music]) {
       expect(factory).toHaveBeenCalledWith(
-        expect.objectContaining({ agentSessionKey: "agent:main:cron:daily-media:run:run-123" }),
+        expect.objectContaining({ agentSessionKey: runSessionKey }),
       );
     }
   });
diff --git a/test/e2e/qa-lab/media/image-generation-lifecycle.e2e.test.ts b/test/e2e/qa-lab/media/image-generation-lifecycle.e2e.test.ts
index ec869f5c51d1..14d85be9e295 100644
--- a/test/e2e/qa-lab/media/image-generation-lifecycle.e2e.test.ts
+++ b/test/e2e/qa-lab/media/image-generation-lifecycle.e2e.test.ts
@@ -21,6 +21,8 @@ const IMAGE_MODEL_REF = "openai/gpt-image-1";
 const REQUEST_TEXT =
   "Image generation check IMAGE_TASK_LIFECYCLE: generate the QA lighthouse image.";
 const CONVERSATION = { id: "image-generation-lifecycle", kind: "direct" as const };
+// QA direct chats share the agent main session, which owns the media task.
+const REQUESTER_SESSION_KEY = "agent:qa:main";
 
 async function readRequestBody(request: IncomingMessage) {
   const chunks: Buffer[] = [];
@@ -156,6 +158,17 @@ async function waitForToolOutput(baseUrl: string, needle: string) {
   return matched as MockOpenAiRequestSnapshot;
 }
 
+async function readImageTaskStatus(gateway: {
+  call: (method: string, params: unknown) => Promise<unknown>;
+}) {
+  const result = (await gateway.call("tools.invoke", {
+    name: "image_generate",
+    args: { action: "status" },
+    sessionKey: REQUESTER_SESSION_KEY,
+  })) as { ok?: boolean; output?: unknown; result?: unknown };
+  return JSON.stringify(result);
+}
+
 describe("image generation task lifecycle through QA-channel", () => {
   const cleanups: Array<() => Promise<void>> = [];
 
@@ -225,6 +238,7 @@ describe("image generation task lifecycle through QA-channel", () => {
     )?.[1];
     expect(runningReceipt).toEqual(expect.any(String));
     expect(imageProvider.requests).toHaveLength(1);
+    expect(await readImageTaskStatus(gateway)).toContain(runningReceipt);
     imageProvider.release();
     await vi.waitFor(
       () => {
@@ -263,6 +277,15 @@ describe("image generation task lifecycle through QA-channel", () => {
     );
     expect(completionReentry.allInputText).toContain("sourceChannel=internal");
 
+    // Completion delivery precedes the terminal task state; wait for it before the next duplicate.
+    await vi.waitFor(
+      async () =>
+        expect(await readImageTaskStatus(gateway)).toContain(
+          "No active image generation task is currently running",
+        ),
+      { interval: 50, timeout: 30_000 },
+    );
+
     await sendExactRequest();
     const completedDuplicate = await waitForToolOutput(mock.baseUrl, "recently succeeded");
     expect(completedDuplicate.toolOutput).toContain(runningReceipt);