diff --git a/managed-inference/qualifications/llama-cpp.openclaw.spark-single.v1.yaml b/managed-inference/qualifications/llama-cpp.openclaw.spark-single.v1.yaml
index 2e1751e4e3f..55a732989f3 100644
--- a/managed-inference/qualifications/llama-cpp.openclaw.spark-single.v1.yaml
+++ b/managed-inference/qualifications/llama-cpp.openclaw.spark-single.v1.yaml
@@ -8,7 +8,7 @@ metadata:
   id: llama-cpp.openclaw.spark-single.v1
 
 spec:
-  execution: disabled
+  execution: enabled
   agent: openclaw
   image:
     reference: ghcr.io/nvidia/nemoclaw/openclaw-sandbox@sha256:3648441718cdd6c2bc4c8fe39fa0d04d3931656b2063af34215cc51841cd0d5e
diff --git a/test/llama-cpp-dgx-spark-qualification-plan.test.ts b/test/llama-cpp-dgx-spark-qualification-plan.test.ts
index 145cd9c92a3..407501c2580 100644
--- a/test/llama-cpp-dgx-spark-qualification-plan.test.ts
+++ b/test/llama-cpp-dgx-spark-qualification-plan.test.ts
@@ -98,7 +98,7 @@ describe("llama.cpp DGX Spark qualification plan export (#8260)", () => {
     );
 
     expect(output.execution).toBe("enabled");
-    expect(output.agent_qualification_execution).toBe("disabled");
+    expect(output.agent_qualification_execution).toBe("enabled");
     expect(output.runner).toBe("linux-arm64-gpu-dgx-spark-gb10-protected-1");
     expect(parseLlamaCppDgxSparkQualificationPlan(JSON.parse(output.qualification))).toMatchObject({
       execution: "enabled",
@@ -112,7 +112,7 @@ describe("llama.cpp DGX Spark qualification plan export (#8260)", () => {
       qualification: {
         agentQualification: {
           agent: "openclaw",
-          execution: "disabled",
+          execution: "enabled",
           runtimeProvider: "docker",
         },
         probeBounds: {
@@ -146,6 +146,7 @@ describe("llama.cpp DGX Spark qualification plan export (#8260)", () => {
     const output = exportLlamaCppDgxSparkQualificationPlan(repoRoot);
 
     expect(output).toMatchObject({
+      agent_qualification_execution: "enabled",
       environment: "approve-dgx-spark-image-qualification",
       execution: "enabled",
       model_host_path: "/var/lib/nemoclaw/models/Nemotron-3-Nano-30B-A3B-UD-Q4_K_XL.gguf",
diff --git a/test/llama-cpp-image.test.ts b/test/llama-cpp-image.test.ts
index a56c91f51f8..4af431ad512 100644
--- a/test/llama-cpp-image.test.ts
+++ b/test/llama-cpp-image.test.ts
@@ -229,7 +229,7 @@ describe("declarative llama.cpp server image", () => {
       },
       qualification: {
         agentQualification: {
-          execution: "disabled",
+          execution: "enabled",
           image: {
             reference:
               "ghcr.io/nvidia/nemoclaw/openclaw-sandbox@sha256:3648441718cdd6c2bc4c8fe39fa0d04d3931656b2063af34215cc51841cd0d5e",
@@ -445,7 +445,7 @@ describe("declarative llama.cpp server image", () => {
       manifestSource,
       recipeSource,
       undefined,
-      qualificationSource.replace("execution: disabled", "execution: enabled"),
+      qualificationSource,
     );
 
     expect(JSON.parse(enabled.publication_qualification_plan)).toMatchObject({