diff --git a/src/auto-reply/reply/get-reply-inline-actions.tool-policy.test.ts b/src/auto-reply/reply/get-reply-inline-actions.tool-policy.test.ts
index 60f374db5e5..961ba4d4ee3 100644
--- a/src/auto-reply/reply/get-reply-inline-actions.tool-policy.test.ts
+++ b/src/auto-reply/reply/get-reply-inline-actions.tool-policy.test.ts
@@ -43,7 +43,7 @@ const createTypingController = (): TypingController => ({
   cleanup: vi.fn(),
 });
 
-const skillCommands: SkillCommandSpec[] = [
+const defaultSkillCommands: SkillCommandSpec[] = [
   {
     name: "danger-skill",
     skillName: "danger-skill",
@@ -53,12 +53,26 @@ const skillCommands: SkillCommandSpec[] = [
       toolName: "gateway",
     },
   },
+  {
+    name: "read-skill",
+    skillName: "read-skill",
+    description: "Allowed direct tool dispatch",
+    dispatch: {
+      kind: "tool",
+      toolName: "read",
+    },
+  },
 ];
 
-function createInput(): HandleInlineActionsInput {
+function createInput(overrides?: {
+  body?: string;
+  senderIsOwner?: boolean;
+  skillCommands?: SkillCommandSpec[];
+}): HandleInlineActionsInput {
+  const body = overrides?.body ?? "/danger-skill test";
   const ctx = buildTestCtx({
-    Body: "/danger-skill test",
-    CommandBody: "/danger-skill test",
+    Body: body,
+    CommandBody: body,
     Provider: "whatsapp",
     Surface: "whatsapp",
     From: "whatsapp:+123",
@@ -85,17 +99,17 @@ function createInput(): HandleInlineActionsInput {
       channel: "whatsapp",
       channelId: "whatsapp",
       ownerList: [],
-      senderIsOwner: true,
+      senderIsOwner: overrides?.senderIsOwner ?? true,
       isAuthorizedSender: true,
       senderId: "owner-1",
       abortKey: "whatsapp:+123",
-      rawBodyNormalized: "/danger-skill test",
-      commandBodyNormalized: "/danger-skill test",
+      rawBodyNormalized: body,
+      commandBodyNormalized: body,
       from: "whatsapp:+123",
       to: "whatsapp:+123",
     },
-    directives: clearInlineDirectives("/danger-skill test"),
-    cleanedBody: "/danger-skill test",
+    directives: clearInlineDirectives(body),
+    cleanedBody: body,
     elevatedEnabled: false,
     elevatedAllowed: false,
     elevatedFailures: [],
@@ -110,7 +124,7 @@ function createInput(): HandleInlineActionsInput {
     contextTokens: 0,
     abortedLastRun: false,
     sessionScope: "per-sender",
-    skillCommands,
+    skillCommands: overrides?.skillCommands ?? defaultSkillCommands,
   };
 }
 
@@ -130,4 +144,24 @@ describe("handleInlineActions skill tool dispatch", () => {
     });
     expect(gatewayExecuteMock).not.toHaveBeenCalled();
   });
+
+  it("executes an allowed tool through direct /skill dispatch", async () => {
+    const result = await handleInlineActions(createInput({ body: "/read-skill test" }));
+
+    expect(result).toEqual({
+      kind: "reply",
+      reply: { text: "READ" },
+    });
+    expect(readExecuteMock).toHaveBeenCalled();
+  });
+
+  it("keeps owner-only tools blocked for non-owners before policy resolution", async () => {
+    const result = await handleInlineActions(createInput({ senderIsOwner: false }));
+
+    expect(result).toEqual({
+      kind: "reply",
+      reply: { text: "❌ Tool not available: gateway" },
+    });
+    expect(gatewayExecuteMock).not.toHaveBeenCalled();
+  });
 });
diff --git a/src/auto-reply/reply/get-reply-inline-actions.ts b/src/auto-reply/reply/get-reply-inline-actions.ts
index cc2cb4d0548..9529bbe5317 100644
--- a/src/auto-reply/reply/get-reply-inline-actions.ts
+++ b/src/auto-reply/reply/get-reply-inline-actions.ts
@@ -127,7 +127,7 @@ function resolveSkillDispatchTools(params: {
   });
   const toolsByAuthorization = applyOwnerOnlyToolPolicy(tools, params.senderIsOwner);
   const {
-    agentId,
+    agentId: resolvedAgentId,
     globalPolicy,
     globalProviderPolicy,
     agentPolicy,
@@ -194,7 +194,7 @@ function resolveSkillDispatchTools(params: {
         agentPolicy,
         agentProviderPolicy,
         groupPolicy,
-        agentId,
+        agentId: resolvedAgentId,
       }),
       { policy: sandboxPolicy, label: "sandbox tools.allow" },
       { policy: subagentPolicy, label: "subagent tools.allow" },