diff --git a/scripts/github/real-behavior-proof-policy.mjs b/scripts/github/real-behavior-proof-policy.mjs
index c76ab5b9bf1..b4242a91276 100644
--- a/scripts/github/real-behavior-proof-policy.mjs
+++ b/scripts/github/real-behavior-proof-policy.mjs
@@ -6,6 +6,7 @@ export const MOCK_ONLY_PROOF_LABEL = "triage: mock-only-proof";
 export const MAINTAINER_TEAM_SLUG = "maintainer";
 
 export const CLAWSWEEPER_PROOF_VERDICT_STATUS = "clawsweeper_exact_head_pass";
+const CLAWSWEEPER_BOT_LOGIN = "clawsweeper[bot]";
 
 const privilegedAuthorAssociations = new Set(["OWNER", "MEMBER", "COLLABORATOR"]);
 
@@ -246,7 +247,14 @@ function isTrustedClawSweeperComment(comment) {
   const appSlug = String(
     comment?.performed_via_github_app?.slug ?? comment?.performedViaGithubApp?.slug ?? "",
   ).toLowerCase();
-  return appSlug === "clawsweeper";
+  if (appSlug === "clawsweeper") {
+    return true;
+  }
+  // GitHub can omit performed_via_github_app on issue comments while still
+  // returning the reserved App bot identity.
+  const login = String(comment?.user?.login ?? "").toLowerCase();
+  const userType = String(comment?.user?.type ?? "");
+  return login === CLAWSWEEPER_BOT_LOGIN && userType === "Bot";
 }
 
 export function hasClawSweeperExactHeadProof({ pullRequest, comments = [] } = {}) {
diff --git a/src/cron/isolated-agent.model-overrides.test.ts b/src/cron/isolated-agent.model-overrides.test.ts
index 0cf67de28ac..7d3d889c0ae 100644
--- a/src/cron/isolated-agent.model-overrides.test.ts
+++ b/src/cron/isolated-agent.model-overrides.test.ts
@@ -1,4 +1,5 @@
 import "./isolated-agent.mocks.js";
+import path from "node:path";
 import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import { loadModelCatalog } from "../agents/model-catalog.js";
 import { runEmbeddedPiAgent } from "../agents/pi-embedded.js";
@@ -21,7 +22,7 @@ import * as isolatedAgentRunRuntime from "./isolated-agent/run.runtime.js";
 
 function installThinkingTestProviders() {
   const registry = createTestRegistry();
-  registry.providers = ["anthropic", "openai", "openrouter"].map(
+  registry.providers = ["anthropic", "google", "openai", "openrouter"].map(
     (providerId): PluginProviderRegistration => ({
       pluginId: providerId,
       source: "test",
@@ -29,10 +30,18 @@ function installThinkingTestProviders() {
         id: providerId,
         label: providerId,
         auth: [],
-        resolveThinkingProfile: () => ({
-          levels: BASE_THINKING_LEVELS.map((id) => ({ id })),
-          defaultLevel: "off",
-        }),
+        resolveThinkingProfile: ({ modelId }) =>
+          providerId === "google" && modelId === "gemini-3-flash-preview"
+            ? {
+                levels: (["off", "minimal", "low", "medium", "adaptive", "high"] as const).map(
+                  (id) => ({ id }),
+                ),
+                preserveWhenCatalogReasoningFalse: true,
+              }
+            : {
+                levels: BASE_THINKING_LEVELS.map((id) => ({ id })),
+                defaultLevel: "off",
+              },
       },
     }),
   );
@@ -253,4 +262,38 @@ describe("runCronIsolatedAgentTurn model overrides", () => {
       expect(callArgs?.thinkLevel).toBe("low");
     });
   });
+
+  it("keeps configured Gemini 3 cron thinking when catalog reasoning metadata is stale", async () => {
+    await withTempHome(async (home) => {
+      vi.mocked(isolatedAgentRunRuntime.resolveThinkingDefault).mockReturnValueOnce("low");
+      vi.mocked(loadModelCatalog).mockResolvedValueOnce([
+        {
+          id: "gemini-3-flash-preview",
+          name: "Gemini 3 Flash Preview",
+          provider: "google",
+          reasoning: false,
+        },
+      ]);
+
+      await runCronTurn(home, {
+        cfgOverrides: {
+          agents: {
+            defaults: {
+              model: "google/gemini-3-flash-preview",
+              workspace: path.join(home, "openclaw"),
+              thinkingDefault: "low",
+            },
+          },
+        },
+        jobPayload: DEFAULT_AGENT_TURN_PAYLOAD,
+        mockTexts: ["done"],
+      });
+
+      const calls = vi.mocked(runEmbeddedPiAgent).mock.calls;
+      const callArgs = calls[calls.length - 1]?.[0];
+      expect(callArgs?.provider).toBe("google");
+      expect(callArgs?.model).toBe("gemini-3-flash-preview");
+      expect(callArgs?.thinkLevel).toBe("low");
+    });
+  });
 });
diff --git a/test/scripts/real-behavior-proof-policy.test.ts b/test/scripts/real-behavior-proof-policy.test.ts
index 01286df72cb..7ce18a4b24f 100644
--- a/test/scripts/real-behavior-proof-policy.test.ts
+++ b/test/scripts/real-behavior-proof-policy.test.ts
@@ -234,7 +234,7 @@ describe("real-behavior-proof-policy", () => {
     expect(evaluateClawSweeperExactHeadProof({ pullRequest, comments }).passed).toBe(false);
   });
 
-  it("rejects bot-shaped ClawSweeper pass verdict markers without the GitHub App source", () => {
+  it("accepts exact ClawSweeper bot pass verdict markers when GitHub omits the app source", () => {
     const pullRequest = {
       number: 83581,
       head: {
@@ -251,6 +251,27 @@ describe("real-behavior-proof-policy", () => {
       },
     ];
 
+    expect(hasClawSweeperExactHeadProof({ pullRequest, comments })).toBe(true);
+    expect(evaluateClawSweeperExactHeadProof({ pullRequest, comments }).passed).toBe(true);
+  });
+
+  it("rejects bot-shaped pass verdict markers from other bot users", () => {
+    const pullRequest = {
+      number: 83581,
+      head: {
+        sha: "06ee95df6608d29a395c52ba8ab53fdd93a9dc4f",
+      },
+    };
+    const comments = [
+      {
+        user: {
+          login: "not-clawsweeper[bot]",
+          type: "Bot",
+        },
+        body: "<!-- clawsweeper-verdict:pass item=83581 sha=06ee95df6608d29a395c52ba8ab53fdd93a9dc4f confidence=high -->",
+      },
+    ];
+
     expect(hasClawSweeperExactHeadProof({ pullRequest, comments })).toBe(false);
     expect(evaluateClawSweeperExactHeadProof({ pullRequest, comments }).passed).toBe(false);
   });