test(auto-reply): dedupe directive behavior e2e fixtures

2026-02-19 18:39:20 -05:00 · 2026-02-18 12:20:40 +00:00
parent 3c886ee98b
commit 2fd211b705
6 changed files with 185 additions and 348 deletions
--- a/src/auto-reply/reply.directive.directive-behavior.defaults-think-low-reasoning-capable-models-no.e2e.test.ts
+++ b/src/auto-reply/reply.directive.directive-behavior.defaults-think-low-reasoning-capable-models-no.e2e.test.ts
@@ -1,48 +1,20 @@
 import "./reply.directive.directive-behavior.e2e-mocks.js";
-import path from "node:path";
 import { describe, expect, it, vi } from "vitest";
-import type { OpenClawConfig } from "../config/config.js";
 import {
  installDirectiveBehaviorE2EHooks,
  loadModelCatalog,
+  makeEmbeddedTextResult,
+  makeWhatsAppDirectiveConfig,
+  mockEmbeddedTextResult,
+  replyText,
+  replyTexts,
  runEmbeddedPiAgent,
  withTempHome,
 } from "./reply.directive.directive-behavior.e2e-harness.js";
 import { getReplyFromConfig } from "./reply.js";

-function makeThinkConfig(home: string) {
-  return {
-    agents: {
-      defaults: {
-        model: "anthropic/claude-opus-4-5",
-        workspace: path.join(home, "openclaw"),
-      },
-    },
-    session: { store: path.join(home, "sessions.json") },
-  } as unknown as OpenClawConfig;
-}
-
-function makeWhatsAppConfig(home: string) {
-  return {
-    agents: {
-      defaults: {
-        model: "anthropic/claude-opus-4-5",
-        workspace: path.join(home, "openclaw"),
-      },
-    },
-    channels: { whatsapp: { allowFrom: ["*"] } },
-    session: { store: path.join(home, "sessions.json") },
-  } as unknown as OpenClawConfig;
-}
-
 async function runReplyToCurrentCase(home: string, text: string) {
-  vi.mocked(runEmbeddedPiAgent).mockResolvedValue({
-    payloads: [{ text }],
-    meta: {
-      durationMs: 5,
-      agentMeta: { sessionId: "s", provider: "p", model: "m" },
-    },
-  });
+  vi.mocked(runEmbeddedPiAgent).mockResolvedValue(makeEmbeddedTextResult(text));

  const res = await getReplyFromConfig(
    {
@@ -52,7 +24,7 @@ async function runReplyToCurrentCase(home: string, text: string) {
      MessageSid: "msg-123",
    },
    {},
-    makeWhatsAppConfig(home),
+    makeWhatsAppDirectiveConfig(home, { model: "anthropic/claude-opus-4-5" }),
  );

  return Array.isArray(res) ? res[0] : res;
@@ -75,10 +47,10 @@ describe("directive behavior", () => {
      const res = await getReplyFromConfig(
        { Body: "/think", From: "+1222", To: "+1222", CommandAuthorized: true },
        {},
-        makeThinkConfig(home),
+        makeWhatsAppDirectiveConfig(home, { model: "anthropic/claude-opus-4-5" }),
      );

-      const text = Array.isArray(res) ? res[0]?.text : res?.text;
+      const text = replyText(res);
      expect(text).toContain("Current thinking level: low");
      expect(text).toContain("Options: off, minimal, low, medium, high.");
      expect(runEmbeddedPiAgent).not.toHaveBeenCalled();
@@ -98,42 +70,29 @@ describe("directive behavior", () => {
      const res = await getReplyFromConfig(
        { Body: "/think", From: "+1222", To: "+1222", CommandAuthorized: true },
        {},
-        makeThinkConfig(home),
+        makeWhatsAppDirectiveConfig(home, { model: "anthropic/claude-opus-4-5" }),
      );

-      const text = Array.isArray(res) ? res[0]?.text : res?.text;
+      const text = replyText(res);
      expect(text).toContain("Current thinking level: off");
      expect(text).toContain("Options: off, minimal, low, medium, high.");
      expect(runEmbeddedPiAgent).not.toHaveBeenCalled();
    });
  });
-  it("strips reply tags and maps reply_to_current to MessageSid", async () => {
-    await withTempHome(async (home) => {
-      const payload = await runReplyToCurrentCase(home, "hello [[reply_to_current]]");
-      expect(payload?.text).toBe("hello");
-      expect(payload?.replyToId).toBe("msg-123");
+  for (const replyTag of ["[[reply_to_current]]", "[[ reply_to_current ]]"]) {
+    it(`strips ${replyTag} and maps reply_to_current to MessageSid`, async () => {
+      await withTempHome(async (home) => {
+        const payload = await runReplyToCurrentCase(home, `hello ${replyTag}`);
+        expect(payload?.text).toBe("hello");
+        expect(payload?.replyToId).toBe("msg-123");
+      });
    });
-  });
-  it("strips reply tags with whitespace and maps reply_to_current to MessageSid", async () => {
-    await withTempHome(async (home) => {
-      const payload = await runReplyToCurrentCase(home, "hello [[ reply_to_current ]]");
-      expect(payload?.text).toBe("hello");
-      expect(payload?.replyToId).toBe("msg-123");
-    });
-  });
+  }
  it("prefers explicit reply_to id over reply_to_current", async () => {
    await withTempHome(async (home) => {
-      vi.mocked(runEmbeddedPiAgent).mockResolvedValue({
-        payloads: [
-          {
-            text: "hi [[reply_to_current]] [[reply_to:abc-456]]",
-          },
-        ],
-        meta: {
-          durationMs: 5,
-          agentMeta: { sessionId: "s", provider: "p", model: "m" },
-        },
-      });
+      vi.mocked(runEmbeddedPiAgent).mockResolvedValue(
+        makeEmbeddedTextResult("hi [[reply_to_current]] [[reply_to:abc-456]]"),
+      );

      const res = await getReplyFromConfig(
        {
@@ -143,16 +102,7 @@ describe("directive behavior", () => {
          MessageSid: "msg-123",
        },
        {},
-        {
-          agents: {
-            defaults: {
-              model: { primary: "anthropic/claude-opus-4-5" },
-              workspace: path.join(home, "openclaw"),
-            },
-          },
-          channels: { whatsapp: { allowFrom: ["*"] } },
-          session: { store: path.join(home, "sessions.json") },
-        },
+        makeWhatsAppDirectiveConfig(home, { model: { primary: "anthropic/claude-opus-4-5" } }),
      );

      const payload = Array.isArray(res) ? res[0] : res;
@@ -162,13 +112,7 @@ describe("directive behavior", () => {
  });
  it("applies inline think and still runs agent content", async () => {
    await withTempHome(async (home) => {
-      vi.mocked(runEmbeddedPiAgent).mockResolvedValue({
-        payloads: [{ text: "done" }],
-        meta: {
-          durationMs: 5,
-          agentMeta: { sessionId: "s", provider: "p", model: "m" },
-        },
-      });
+      mockEmbeddedTextResult("done");

      const res = await getReplyFromConfig(
        {
@@ -177,19 +121,10 @@ describe("directive behavior", () => {
          To: "+2000",
        },
        {},
-        {
-          agents: {
-            defaults: {
-              model: { primary: "anthropic/claude-opus-4-5" },
-              workspace: path.join(home, "openclaw"),
-            },
-          },
-          channels: { whatsapp: { allowFrom: ["*"] } },
-          session: { store: path.join(home, "sessions.json") },
-        },
+        makeWhatsAppDirectiveConfig(home, { model: { primary: "anthropic/claude-opus-4-5" } }),
      );

-      const texts = (Array.isArray(res) ? res : [res]).map((entry) => entry?.text).filter(Boolean);
+      const texts = replyTexts(res);
      expect(texts).toContain("done");
      expect(runEmbeddedPiAgent).toHaveBeenCalledOnce();
    });
--- a/src/auto-reply/reply.directive.directive-behavior.e2e-harness.ts
+++ b/src/auto-reply/reply.directive.directive-behavior.e2e-harness.ts
@@ -36,6 +36,20 @@ export function replyTexts(res: ReplyPayloadText | ReplyPayloadText[]): string[]
    .filter((value): value is string => Boolean(value));
 }

+export function makeEmbeddedTextResult(text = "done") {
+  return {
+    payloads: [{ text }],
+    meta: {
+      durationMs: 5,
+      agentMeta: { sessionId: "s", provider: "p", model: "m" },
+    },
+  };
+}
+
+export function mockEmbeddedTextResult(text = "done") {
+  vi.mocked(runEmbeddedPiAgent).mockResolvedValue(makeEmbeddedTextResult(text));
+}
+
 export async function withTempHome<T>(fn: (home: string) => Promise<T>): Promise<T> {
  return withTempHomeBase(
    async (home) => {
--- a/src/auto-reply/reply.directive.directive-behavior.ignores-inline-model-uses-default-model.e2e.test.ts
+++ b/src/auto-reply/reply.directive.directive-behavior.ignores-inline-model-uses-default-model.e2e.test.ts
@@ -1,27 +1,32 @@
 import "./reply.directive.directive-behavior.e2e-mocks.js";
-import path from "node:path";
 import { describe, expect, it, vi } from "vitest";
 import {
  installDirectiveBehaviorE2EHooks,
  loadModelCatalog,
+  makeWhatsAppDirectiveConfig,
+  mockEmbeddedTextResult,
+  replyTexts,
  runEmbeddedPiAgent,
  withTempHome,
 } from "./reply.directive.directive-behavior.e2e-harness.js";
 import { getReplyFromConfig } from "./reply.js";

+function makeDefaultModelConfig(home: string) {
+  return makeWhatsAppDirectiveConfig(home, {
+    model: { primary: "anthropic/claude-opus-4-5" },
+    models: {
+      "anthropic/claude-opus-4-5": {},
+      "openai/gpt-4.1-mini": {},
+    },
+  });
+}
+
 describe("directive behavior", () => {
  installDirectiveBehaviorE2EHooks();

  it("ignores inline /model and uses the default model", async () => {
    await withTempHome(async (home) => {
-      const storePath = path.join(home, "sessions.json");
-      vi.mocked(runEmbeddedPiAgent).mockResolvedValue({
-        payloads: [{ text: "done" }],
-        meta: {
-          durationMs: 5,
-          agentMeta: { sessionId: "s", provider: "p", model: "m" },
-        },
-      });
+      mockEmbeddedTextResult("done");

      const res = await getReplyFromConfig(
        {
@@ -30,23 +35,10 @@ describe("directive behavior", () => {
          To: "+2000",
        },
        {},
-        {
-          agents: {
-            defaults: {
-              model: { primary: "anthropic/claude-opus-4-5" },
-              workspace: path.join(home, "openclaw"),
-              models: {
-                "anthropic/claude-opus-4-5": {},
-                "openai/gpt-4.1-mini": {},
-              },
-            },
-          },
-          channels: { whatsapp: { allowFrom: ["*"] } },
-          session: { store: storePath },
-        },
+        makeDefaultModelConfig(home),
      );

-      const texts = (Array.isArray(res) ? res : [res]).map((entry) => entry?.text).filter(Boolean);
+      const texts = replyTexts(res);
      expect(texts).toContain("done");
      expect(runEmbeddedPiAgent).toHaveBeenCalledOnce();
      const call = vi.mocked(runEmbeddedPiAgent).mock.calls[0]?.[0];
@@ -56,14 +48,7 @@ describe("directive behavior", () => {
  });
  it("defaults thinking to low for reasoning-capable models", async () => {
    await withTempHome(async (home) => {
-      const storePath = path.join(home, "sessions.json");
-      vi.mocked(runEmbeddedPiAgent).mockResolvedValue({
-        payloads: [{ text: "done" }],
-        meta: {
-          durationMs: 5,
-          agentMeta: { sessionId: "s", provider: "p", model: "m" },
-        },
-      });
+      mockEmbeddedTextResult("done");
      vi.mocked(loadModelCatalog).mockResolvedValueOnce([
        {
          id: "claude-opus-4-5",
@@ -80,16 +65,7 @@ describe("directive behavior", () => {
          To: "+2000",
        },
        {},
-        {
-          agents: {
-            defaults: {
-              model: { primary: "anthropic/claude-opus-4-5" },
-              workspace: path.join(home, "openclaw"),
-            },
-          },
-          channels: { whatsapp: { allowFrom: ["*"] } },
-          session: { store: storePath },
-        },
+        makeWhatsAppDirectiveConfig(home, { model: { primary: "anthropic/claude-opus-4-5" } }),
      );

      expect(runEmbeddedPiAgent).toHaveBeenCalledOnce();
@@ -99,14 +75,7 @@ describe("directive behavior", () => {
  });
  it("passes elevated defaults when sender is approved", async () => {
    await withTempHome(async (home) => {
-      const storePath = path.join(home, "sessions.json");
-      vi.mocked(runEmbeddedPiAgent).mockResolvedValue({
-        payloads: [{ text: "done" }],
-        meta: {
-          durationMs: 5,
-          agentMeta: { sessionId: "s", provider: "p", model: "m" },
-        },
-      });
+      mockEmbeddedTextResult("done");

      await getReplyFromConfig(
        {
@@ -117,21 +86,17 @@ describe("directive behavior", () => {
          SenderE164: "+1004",
        },
        {},
-        {
-          agents: {
-            defaults: {
-              model: { primary: "anthropic/claude-opus-4-5" },
-              workspace: path.join(home, "openclaw"),
+        makeWhatsAppDirectiveConfig(
+          home,
+          { model: { primary: "anthropic/claude-opus-4-5" } },
+          {
+            tools: {
+              elevated: {
+                allowFrom: { whatsapp: ["+1004"] },
+              },
            },
          },
-          tools: {
-            elevated: {
-              allowFrom: { whatsapp: ["+1004"] },
-            },
-          },
-          channels: { whatsapp: { allowFrom: ["*"] } },
-          session: { store: storePath },
-        },
+        ),
      );

      expect(runEmbeddedPiAgent).toHaveBeenCalledOnce();
--- a/src/auto-reply/reply.directive.directive-behavior.prefers-alias-matches-fuzzy-selection-is-ambiguous.e2e.test.ts
+++ b/src/auto-reply/reply.directive.directive-behavior.prefers-alias-matches-fuzzy-selection-is-ambiguous.e2e.test.ts
@@ -9,7 +9,10 @@ import {
  assertModelSelection,
  installDirectiveBehaviorE2EHooks,
  MAIN_SESSION_KEY,
+  makeWhatsAppDirectiveConfig,
+  replyText,
  runEmbeddedPiAgent,
+  sessionStorePath,
  withTempHome,
 } from "./reply.directive.directive-behavior.e2e-harness.js";
 import { getReplyFromConfig } from "./reply.js";
@@ -26,12 +29,22 @@ function makeModelDefinition(id: string, name: string): ModelDefinitionConfig {
  };
 }

+function makeModelSwitchConfig(home: string) {
+  return makeWhatsAppDirectiveConfig(home, {
+    model: { primary: "openai/gpt-4.1-mini" },
+    models: {
+      "openai/gpt-4.1-mini": {},
+      "anthropic/claude-opus-4-5": { alias: "Opus" },
+    },
+  });
+}
+
 describe("directive behavior", () => {
  installDirectiveBehaviorE2EHooks();

  it("prefers alias matches when fuzzy selection is ambiguous", async () => {
    await withTempHome(async (home) => {
-      const storePath = path.join(home, "sessions.json");
+      const storePath = sessionStorePath(home);

      const res = await getReplyFromConfig(
        { Body: "/model ki", From: "+1222", To: "+1222", CommandAuthorized: true },
@@ -69,7 +82,7 @@ describe("directive behavior", () => {
        },
      );

-      const text = Array.isArray(res) ? res[0]?.text : res?.text;
+      const text = replyText(res);
      expect(text).toContain("Model set to Kimi (moonshot/kimi-k2-0905-preview).");
      assertModelSelection(storePath, {
        provider: "moonshot",
@@ -80,7 +93,7 @@ describe("directive behavior", () => {
  });
  it("stores auth profile overrides on /model directive", async () => {
    await withTempHome(async (home) => {
-      const storePath = path.join(home, "sessions.json");
+      const storePath = sessionStorePath(home);
      const authDir = path.join(home, ".openclaw", "agents", "main", "agent");
      await fs.mkdir(authDir, { recursive: true, mode: 0o700 });
      await fs.writeFile(
@@ -104,22 +117,10 @@ describe("directive behavior", () => {
      const res = await getReplyFromConfig(
        { Body: "/model Opus@anthropic:work", From: "+1222", To: "+1222", CommandAuthorized: true },
        {},
-        {
-          agents: {
-            defaults: {
-              model: { primary: "openai/gpt-4.1-mini" },
-              workspace: path.join(home, "openclaw"),
-              models: {
-                "openai/gpt-4.1-mini": {},
-                "anthropic/claude-opus-4-5": { alias: "Opus" },
-              },
-            },
-          },
-          session: { store: storePath },
-        },
+        makeModelSwitchConfig(home),
      );

-      const text = Array.isArray(res) ? res[0]?.text : res?.text;
+      const text = replyText(res);
      expect(text).toContain("Auth profile set to anthropic:work");
      const store = loadSessionStore(storePath);
      const entry = store["agent:main:main"];
@@ -130,24 +131,10 @@ describe("directive behavior", () => {
  it("queues a system event when switching models", async () => {
    await withTempHome(async (home) => {
      drainSystemEvents(MAIN_SESSION_KEY);
-      const storePath = path.join(home, "sessions.json");
-
      await getReplyFromConfig(
        { Body: "/model Opus", From: "+1222", To: "+1222", CommandAuthorized: true },
        {},
-        {
-          agents: {
-            defaults: {
-              model: { primary: "openai/gpt-4.1-mini" },
-              workspace: path.join(home, "openclaw"),
-              models: {
-                "openai/gpt-4.1-mini": {},
-                "anthropic/claude-opus-4-5": { alias: "Opus" },
-              },
-            },
-          },
-          session: { store: storePath },
-        },
+        makeModelSwitchConfig(home),
      );

      const events = drainSystemEvents(MAIN_SESSION_KEY);
@@ -158,7 +145,6 @@ describe("directive behavior", () => {
  it("queues a system event when toggling elevated", async () => {
    await withTempHome(async (home) => {
      drainSystemEvents(MAIN_SESSION_KEY);
-      const storePath = path.join(home, "sessions.json");

      await getReplyFromConfig(
        {
@@ -169,17 +155,11 @@ describe("directive behavior", () => {
          CommandAuthorized: true,
        },
        {},
-        {
-          agents: {
-            defaults: {
-              model: { primary: "openai/gpt-4.1-mini" },
-              workspace: path.join(home, "openclaw"),
-            },
-          },
-          tools: { elevated: { allowFrom: { whatsapp: ["*"] } } },
-          channels: { whatsapp: { allowFrom: ["*"] } },
-          session: { store: storePath },
-        },
+        makeWhatsAppDirectiveConfig(
+          home,
+          { model: { primary: "openai/gpt-4.1-mini" } },
+          { tools: { elevated: { allowFrom: { whatsapp: ["*"] } } } },
+        ),
      );

      const events = drainSystemEvents(MAIN_SESSION_KEY);
@@ -189,7 +169,6 @@ describe("directive behavior", () => {
  it("queues a system event when toggling reasoning", async () => {
    await withTempHome(async (home) => {
      drainSystemEvents(MAIN_SESSION_KEY);
-      const storePath = path.join(home, "sessions.json");

      await getReplyFromConfig(
        {
@@ -200,16 +179,7 @@ describe("directive behavior", () => {
          CommandAuthorized: true,
        },
        {},
-        {
-          agents: {
-            defaults: {
-              model: { primary: "openai/gpt-4.1-mini" },
-              workspace: path.join(home, "openclaw"),
-            },
-          },
-          channels: { whatsapp: { allowFrom: ["*"] } },
-          session: { store: storePath },
-        },
+        makeWhatsAppDirectiveConfig(home, { model: { primary: "openai/gpt-4.1-mini" } }),
      );

      const events = drainSystemEvents(MAIN_SESSION_KEY);
--- a/src/auto-reply/reply.directive.directive-behavior.requires-per-agent-allowlist-addition-global.e2e.test.ts
+++ b/src/auto-reply/reply.directive.directive-behavior.requires-per-agent-allowlist-addition-global.e2e.test.ts
@@ -42,6 +42,29 @@ function makeWorkElevatedAllowlistConfig(home: string) {
  };
 }

+function makeElevatedDirectiveConfig(
+  home: string,
+  defaults: Record<string, unknown> = {},
+  extra: Record<string, unknown> = {},
+) {
+  return makeWhatsAppDirectiveConfig(
+    home,
+    {
+      model: "anthropic/claude-opus-4-5",
+      ...defaults,
+    },
+    {
+      tools: {
+        elevated: {
+          allowFrom: { whatsapp: ["+1222"] },
+        },
+      },
+      channels: { whatsapp: { allowFrom: ["+1222"] } },
+      ...extra,
+    },
+  );
+}
+
 function makeCommandMessage(body: string, from = "+1222") {
  return {
    Body: body,
@@ -98,21 +121,7 @@ describe("directive behavior", () => {
      const res = await getReplyFromConfig(
        makeCommandMessage("/elevated off"),
        {},
-        makeWhatsAppDirectiveConfig(
-          home,
-          {
-            model: "anthropic/claude-opus-4-5",
-            sandbox: { mode: "off" },
-          },
-          {
-            tools: {
-              elevated: {
-                allowFrom: { whatsapp: ["+1222"] },
-              },
-            },
-            channels: { whatsapp: { allowFrom: ["+1222"] } },
-          },
-        ),
+        makeElevatedDirectiveConfig(home, { sandbox: { mode: "off" } }),
      );

      const text = replyText(res);
@@ -126,18 +135,7 @@ describe("directive behavior", () => {
      const res = await getReplyFromConfig(
        makeCommandMessage("/elevated maybe"),
        {},
-        makeWhatsAppDirectiveConfig(
-          home,
-          { model: "anthropic/claude-opus-4-5" },
-          {
-            tools: {
-              elevated: {
-                allowFrom: { whatsapp: ["+1222"] },
-              },
-            },
-            channels: { whatsapp: { allowFrom: ["+1222"] } },
-          },
-        ),
+        makeElevatedDirectiveConfig(home),
      );

      const text = replyText(res);
@@ -150,18 +148,7 @@ describe("directive behavior", () => {
      const res = await getReplyFromConfig(
        makeCommandMessage("/elevated off\n/verbose on"),
        {},
-        makeWhatsAppDirectiveConfig(
-          home,
-          { model: "anthropic/claude-opus-4-5" },
-          {
-            tools: {
-              elevated: {
-                allowFrom: { whatsapp: ["+1222"] },
-              },
-            },
-            channels: { whatsapp: { allowFrom: ["+1222"] } },
-          },
-        ),
+        makeElevatedDirectiveConfig(home),
      );

      const text = replyText(res);
--- a/src/auto-reply/reply.directive.directive-behavior.updates-tool-verbose-during-flight-run-toggle.e2e.test.ts
+++ b/src/auto-reply/reply.directive.directive-behavior.updates-tool-verbose-during-flight-run-toggle.e2e.test.ts
@@ -3,6 +3,7 @@ import { describe, expect, it, vi } from "vitest";
 import { loadSessionStore, resolveSessionKey, saveSessionStore } from "../config/sessions.js";
 import {
  installDirectiveBehaviorE2EHooks,
+  makeEmbeddedTextResult,
  makeWhatsAppDirectiveConfig,
  replyText,
  replyTexts,
@@ -12,6 +13,62 @@ import {
 } from "./reply.directive.directive-behavior.e2e-harness.js";
 import { getReplyFromConfig } from "./reply.js";

+function makeRunConfig(home: string, storePath: string) {
+  return makeWhatsAppDirectiveConfig(
+    home,
+    { model: "anthropic/claude-opus-4-5" },
+    { session: { store: storePath } },
+  );
+}
+
+async function runInFlightVerboseToggleCase(params: {
+  home: string;
+  shouldEmitBefore: boolean;
+  toggledVerboseLevel: "on" | "off";
+  seedVerboseOn?: boolean;
+}) {
+  const storePath = sessionStorePath(params.home);
+  const ctx = {
+    Body: "please do the thing",
+    From: "+1004",
+    To: "+2000",
+  };
+  const sessionKey = resolveSessionKey(
+    "per-sender",
+    { From: ctx.From, To: ctx.To, Body: ctx.Body },
+    "main",
+  );
+
+  vi.mocked(runEmbeddedPiAgent).mockImplementation(async (agentParams) => {
+    const shouldEmit = agentParams.shouldEmitToolResult;
+    expect(shouldEmit?.()).toBe(params.shouldEmitBefore);
+    const store = loadSessionStore(storePath);
+    const entry = store[sessionKey] ?? {
+      sessionId: "s",
+      updatedAt: Date.now(),
+    };
+    store[sessionKey] = {
+      ...entry,
+      verboseLevel: params.toggledVerboseLevel,
+      updatedAt: Date.now(),
+    };
+    await saveSessionStore(storePath, store);
+    expect(shouldEmit?.()).toBe(!params.shouldEmitBefore);
+    return makeEmbeddedTextResult("done");
+  });
+
+  if (params.seedVerboseOn) {
+    await getReplyFromConfig(
+      { Body: "/verbose on", From: ctx.From, To: ctx.To, CommandAuthorized: true },
+      {},
+      makeRunConfig(params.home, storePath),
+    );
+  }
+
+  const res = await getReplyFromConfig(ctx, {}, makeRunConfig(params.home, storePath));
+  return { res };
+}
+
 async function runModelDirectiveAndGetText(
  home: string,
  body: string,
@@ -35,50 +92,12 @@ describe("directive behavior", () => {

  it("updates tool verbose during an in-flight run (toggle on)", async () => {
    await withTempHome(async (home) => {
-      const storePath = sessionStorePath(home);
-      const ctx = { Body: "please do the thing", From: "+1004", To: "+2000" };
-      const sessionKey = resolveSessionKey(
-        "per-sender",
-        { From: ctx.From, To: ctx.To, Body: ctx.Body },
-        "main",
-      );
-
-      vi.mocked(runEmbeddedPiAgent).mockImplementation(async (params) => {
-        const shouldEmit = params.shouldEmitToolResult;
-        expect(shouldEmit?.()).toBe(false);
-        const store = loadSessionStore(storePath);
-        const entry = store[sessionKey] ?? {
-          sessionId: "s",
-          updatedAt: Date.now(),
-        };
-        store[sessionKey] = {
-          ...entry,
-          verboseLevel: "on",
-          updatedAt: Date.now(),
-        };
-        await saveSessionStore(storePath, store);
-        expect(shouldEmit?.()).toBe(true);
-        return {
-          payloads: [{ text: "done" }],
-          meta: {
-            durationMs: 5,
-            agentMeta: { sessionId: "s", provider: "p", model: "m" },
-          },
-        };
+      const { res } = await runInFlightVerboseToggleCase({
+        home,
+        shouldEmitBefore: false,
+        toggledVerboseLevel: "on",
      });

-      const res = await getReplyFromConfig(
-        ctx,
-        {},
-        makeWhatsAppDirectiveConfig(
-          home,
-          { model: "anthropic/claude-opus-4-5" },
-          {
-            session: { store: storePath },
-          },
-        ),
-      );
-
      const texts = replyTexts(res);
      expect(texts).toContain("done");
      expect(runEmbeddedPiAgent).toHaveBeenCalledOnce();
@@ -86,66 +105,13 @@ describe("directive behavior", () => {
  });
  it("updates tool verbose during an in-flight run (toggle off)", async () => {
    await withTempHome(async (home) => {
-      const storePath = sessionStorePath(home);
-      const ctx = {
-        Body: "please do the thing",
-        From: "+1004",
-        To: "+2000",
-      };
-      const sessionKey = resolveSessionKey(
-        "per-sender",
-        { From: ctx.From, To: ctx.To, Body: ctx.Body },
-        "main",
-      );
-
-      vi.mocked(runEmbeddedPiAgent).mockImplementation(async (params) => {
-        const shouldEmit = params.shouldEmitToolResult;
-        expect(shouldEmit?.()).toBe(true);
-        const store = loadSessionStore(storePath);
-        const entry = store[sessionKey] ?? {
-          sessionId: "s",
-          updatedAt: Date.now(),
-        };
-        store[sessionKey] = {
-          ...entry,
-          verboseLevel: "off",
-          updatedAt: Date.now(),
-        };
-        await saveSessionStore(storePath, store);
-        expect(shouldEmit?.()).toBe(false);
-        return {
-          payloads: [{ text: "done" }],
-          meta: {
-            durationMs: 5,
-            agentMeta: { sessionId: "s", provider: "p", model: "m" },
-          },
-        };
+      const { res } = await runInFlightVerboseToggleCase({
+        home,
+        shouldEmitBefore: true,
+        toggledVerboseLevel: "off",
+        seedVerboseOn: true,
      });

-      await getReplyFromConfig(
-        { Body: "/verbose on", From: ctx.From, To: ctx.To, CommandAuthorized: true },
-        {},
-        makeWhatsAppDirectiveConfig(
-          home,
-          { model: "anthropic/claude-opus-4-5" },
-          {
-            session: { store: storePath },
-          },
-        ),
-      );
-
-      const res = await getReplyFromConfig(
-        ctx,
-        {},
-        makeWhatsAppDirectiveConfig(
-          home,
-          { model: "anthropic/claude-opus-4-5" },
-          {
-            session: { store: storePath },
-          },
-        ),
-      );
-
      const texts = replyTexts(res);
      expect(texts).toContain("done");
      expect(runEmbeddedPiAgent).toHaveBeenCalledOnce();