diff --git a/src/plugin-config.test.ts b/src/plugin-config.test.ts
index 7626ed2f4..8ac0bdee5 100644
--- a/src/plugin-config.test.ts
+++ b/src/plugin-config.test.ts
@@ -3,7 +3,7 @@ import { chmodSync, existsSync, mkdtempSync, mkdirSync, readFileSync, rmSync, wr
 import { tmpdir } from "node:os"
 import { join } from "node:path"
 import * as shared from "./shared"
-import { loadPluginConfig, mergeConfigs, parseConfigPartially } from "./plugin-config";
+import { mergeConfigs, parseConfigPartially } from "./plugin-config";
 import { OhMyOpenCodeConfigSchema, type OhMyOpenCodeConfig } from "./config";
 
 const tempDirs: string[] = []
@@ -12,6 +12,10 @@ function createConfig(config: Partial<OhMyOpenCodeConfig>): OhMyOpenCodeConfig {
   return OhMyOpenCodeConfigSchema.parse(config)
 }
 
+async function importFreshPluginConfigModule(): Promise<typeof import("./plugin-config")> {
+  return import(`./plugin-config?test=${Date.now()}-${Math.random()}`)
+}
+
 afterEach(() => {
   mock.restore()
 
@@ -297,7 +301,7 @@ describe("parseConfigPartially", () => {
 });
 
 describe("loadPluginConfig", () => {
-  it("should only honor mcp_env_allowlist from user config", () => {
+  it("should only honor mcp_env_allowlist from user config", async () => {
     // given
     const rootDir = mkdtempSync(join(tmpdir(), "omo-plugin-config-"))
     const userConfigDir = join(rootDir, "user-config")
@@ -317,16 +321,17 @@ describe("loadPluginConfig", () => {
       JSON.stringify({ mcp_env_allowlist: ["PROJECT_TOKEN"] })
     )
 
-    spyOn(shared, "getOpenCodeConfigDir").mockReturnValue(userConfigDir)
+    process.env.OPENCODE_CONFIG_DIR = userConfigDir
 
     // when
+    const { loadPluginConfig } = await importFreshPluginConfigModule()
     const config = loadPluginConfig(projectDir, {})
 
     // then
     expect(config.mcp_env_allowlist).toEqual(["USER_ONLY_TOKEN"])
   })
 
-  it("should ignore edits to the renamed legacy backup after migration", () => {
+  it("should ignore edits to the renamed legacy backup after migration", async () => {
     // given
     const rootDir = mkdtempSync(join(tmpdir(), "omo-plugin-config-legacy-"))
     const userConfigDir = join(rootDir, "user-config")
@@ -341,9 +346,10 @@ describe("loadPluginConfig", () => {
     mkdirSync(projectConfigDir, { recursive: true })
     writeFileSync(legacyConfigPath, JSON.stringify({ agents: { oracle: { model: "openai/gpt-5.4" } } }))
 
-    spyOn(shared, "getOpenCodeConfigDir").mockReturnValue(userConfigDir)
+    process.env.OPENCODE_CONFIG_DIR = userConfigDir
 
     // when
+    const { loadPluginConfig } = await importFreshPluginConfigModule()
     loadPluginConfig(projectDir, {})
     writeFileSync(backupConfigPath, JSON.stringify({ agents: { oracle: { model: "openai/gpt-5-nano" } } }))
     const reloadedConfig = loadPluginConfig(projectDir, {})
@@ -355,7 +361,7 @@ describe("loadPluginConfig", () => {
     expect(reloadedConfig.agents?.oracle?.model).toBe("openai/gpt-5.4")
   })
 
-  it("should still load config from legacy path when migration fails", () => {
+  it("should still load config from legacy path when migration fails", async () => {
     // given - legacy config exists but canonical path is not writable
     const rootDir = mkdtempSync(join(tmpdir(), "omo-plugin-config-fail-"))
     const userConfigDir = join(rootDir, "user-config")
@@ -374,12 +380,13 @@ describe("loadPluginConfig", () => {
       chmodSync(projectConfigDir, 0o555)
     }
 
-    spyOn(shared, "getOpenCodeConfigDir").mockReturnValue(userConfigDir)
+    process.env.OPENCODE_CONFIG_DIR = userConfigDir
 
     // when
     let config: OhMyOpenCodeConfig
     try {
-      config = loadPluginConfig(projectDir, {})
+      const fresh = await importFreshPluginConfigModule()
+      config = fresh.loadPluginConfig(projectDir, {})
     } finally {
       // Restore permissions for cleanup
       if (process.platform !== "win32") {
@@ -391,7 +398,7 @@ describe("loadPluginConfig", () => {
     expect(config.agents?.oracle?.model).toBe("openai/gpt-5.4")
   })
 
-  it("should load migrated legacy project config on the first load", () => {
+  it("should load migrated legacy project config on the first load", async () => {
     // given
     const rootDir = mkdtempSync(join(tmpdir(), "omo-plugin-config-first-load-"))
     const userConfigDir = join(rootDir, "user-config")
@@ -405,9 +412,10 @@ describe("loadPluginConfig", () => {
     mkdirSync(projectConfigDir, { recursive: true })
     writeFileSync(legacyConfigPath, JSON.stringify({ agents: { oracle: { model: "openai/gpt-5.4" } } }))
 
-    spyOn(shared, "getOpenCodeConfigDir").mockReturnValue(userConfigDir)
+    process.env.OPENCODE_CONFIG_DIR = userConfigDir
 
     // when
+    const { loadPluginConfig } = await importFreshPluginConfigModule()
     const config = loadPluginConfig(projectDir, {})
 
     // then
diff --git a/src/plugin/tool-registry.test.ts b/src/plugin/tool-registry.test.ts
index 4c52b5b79..5c0a42bfb 100644
--- a/src/plugin/tool-registry.test.ts
+++ b/src/plugin/tool-registry.test.ts
@@ -1,4 +1,4 @@
-import { describe, expect, mock, spyOn, test } from "bun:test"
+import { beforeEach, describe, expect, mock, spyOn, test } from "bun:test"
 import { tool } from "@opencode-ai/plugin"
 
 import type { OhMyOpenCodeConfig } from "../config"
@@ -25,7 +25,12 @@ const syncSessionCreatedCallbacks: Array<
   ((event: { sessionID: string; parentID: string; title: string }) => Promise<void>) | undefined
 > = []
 
-mock.module("../tools", () => ({
+const trackedPaneBySession = new Map<string, string>()
+let dispatchOpenClawEvent: ReturnType<typeof spyOn>
+
+const { createToolRegistry, trimToolsToCap } = await import("./tool-registry")
+
+const toolFactories: NonNullable<Parameters<typeof createToolRegistry>[0]["toolFactories"]> = {
   builtinTools: { bash: fakeTool, read: fakeTool },
   createBackgroundTools: mock(() => ({})),
   createCallOmoAgent: mock(() => fakeTool),
@@ -47,12 +52,7 @@ mock.module("../tools", () => ({
   createTaskList: mock(() => fakeTool),
   createTaskUpdateTool: mock(() => fakeTool),
   createHashlineEditTool: mock(() => fakeTool),
-}))
-
-const trackedPaneBySession = new Map<string, string>()
-
-const { createToolRegistry, trimToolsToCap } = await import("./tool-registry")
-const dispatchOpenClawEvent = spyOn(openclawRuntimeDispatch, "dispatchOpenClawEvent")
+}
 
 function createPluginConfig(overrides: Partial<OhMyOpenCodeConfig> = {}): OhMyOpenCodeConfig {
   return {
@@ -65,6 +65,11 @@ function createPluginConfig(overrides: Partial<OhMyOpenCodeConfig> = {}): OhMyOp
   }
 }
 
+beforeEach(() => {
+  dispatchOpenClawEvent = spyOn(openclawRuntimeDispatch, "dispatchOpenClawEvent")
+  syncSessionCreatedCallbacks.length = 0
+})
+
 describe("#given tool trimming prioritization", () => {
   test("#when max_tools trims a hashline edit registration named edit #then edit is removed before higher-priority tools", () => {
     const filteredTools = {
@@ -100,6 +105,7 @@ describe("#given task_system configuration", () => {
         disabledSkills: new Set(),
       },
       availableCategories: [],
+      toolFactories,
     })
 
     expect(result.taskSystemEnabled).toBe(false)
@@ -129,6 +135,7 @@ describe("#given task_system configuration", () => {
         disabledSkills: new Set(),
       },
       availableCategories: [],
+      toolFactories,
     })
 
     expect(result.taskSystemEnabled).toBe(true)
@@ -168,6 +175,7 @@ describe("#given tmux integration is disabled", () => {
       },
       availableCategories: [],
       interactiveBashEnabled: true,
+      toolFactories,
     })
 
     expect(result.filteredTools).toHaveProperty("interactive_bash")
@@ -201,6 +209,7 @@ describe("#given tmux integration is disabled", () => {
       },
       availableCategories: [],
       interactiveBashEnabled: false,
+      toolFactories,
     })
 
     expect(result.filteredTools).not.toHaveProperty("interactive_bash")
@@ -246,6 +255,7 @@ describe("#given openclaw is enabled for sync task sessions", () => {
         disabledSkills: new Set(),
       },
       availableCategories: [],
+      toolFactories,
     })
 
     const onSyncSessionCreated = syncSessionCreatedCallbacks[syncSessionCreatedCallbacks.length - 1]
diff --git a/src/plugin/ultrawork-db-model-override.test.ts b/src/plugin/ultrawork-db-model-override.test.ts
index dc3f15a38..ea5646d26 100644
--- a/src/plugin/ultrawork-db-model-override.test.ts
+++ b/src/plugin/ultrawork-db-model-override.test.ts
@@ -6,6 +6,12 @@ import { tmpdir } from "node:os"
 import * as dataPathModule from "../shared/data-path"
 import * as sharedModule from "../shared"
 
+let scheduleDeferredModelOverride: (typeof import("./ultrawork-db-model-override"))["scheduleDeferredModelOverride"]
+
+async function importFreshUltraworkDbModelOverrideModule(): Promise<typeof import("./ultrawork-db-model-override")> {
+  return import(`./ultrawork-db-model-override?test=${Date.now()}-${Math.random()}`)
+}
+
 function flushMicrotasks(depth: number): Promise<void> {
   return new Promise<void>((resolve) => {
     let remaining = depth
@@ -22,6 +28,11 @@ function flushWithTimeout(): Promise<void> {
   return new Promise<void>((resolve) => setTimeout(resolve, 10))
 }
 
+async function settleDeferredModelOverrideWork(): Promise<void> {
+  await flushMicrotasks(12)
+  await flushWithTimeout()
+}
+
 function isRecord(value: unknown): value is Record<string, unknown> {
   return typeof value === "object" && value !== null
 }
@@ -32,7 +43,7 @@ describe("scheduleDeferredModelOverride", () => {
   let logSpy: ReturnType<typeof spyOn>
   let getDataDirSpy: ReturnType<typeof spyOn>
 
-  beforeEach(() => {
+  beforeEach(async () => {
     tempDir = mkdtempSync(join(tmpdir(), "ultrawork-db-test-"))
     const opencodePath = join(tempDir, "opencode")
     mkdirSync(opencodePath, { recursive: true })
@@ -50,11 +61,15 @@ describe("scheduleDeferredModelOverride", () => {
     `)
     db.close()
 
-    getDataDirSpy = spyOn(dataPathModule, "getDataDir").mockReturnValue(tempDir)
-    logSpy = spyOn(sharedModule, "log").mockImplementation(() => {})
+    getDataDirSpy = spyOn(dataPathModule, "getDataDir")
+    getDataDirSpy.mockReturnValue(tempDir)
+    logSpy = spyOn(sharedModule, "log")
+    logSpy.mockImplementation(() => {})
+    ;({ scheduleDeferredModelOverride } = await importFreshUltraworkDbModelOverrideModule())
   })
 
-  afterEach(() => {
+  afterEach(async () => {
+    await settleDeferredModelOverrideWork()
     getDataDirSpy?.mockRestore()
     logSpy?.mockRestore()
     rmSync(tempDir, { recursive: true, force: true })
@@ -95,7 +110,6 @@ describe("scheduleDeferredModelOverride", () => {
     insertMessage("msg_001", { providerID: "anthropic", modelID: "claude-sonnet-4-6" })
 
     //#when
-    const { scheduleDeferredModelOverride } = await import("./ultrawork-db-model-override")
     scheduleDeferredModelOverride(
       "msg_001",
       { providerID: "anthropic", modelID: "claude-opus-4-6" },
@@ -112,7 +126,6 @@ describe("scheduleDeferredModelOverride", () => {
     insertMessage("msg_002", { providerID: "anthropic", modelID: "claude-sonnet-4-6" })
 
     //#when
-    const { scheduleDeferredModelOverride } = await import("./ultrawork-db-model-override")
     scheduleDeferredModelOverride(
       "msg_002",
       { providerID: "anthropic", modelID: "claude-opus-4-6" },
@@ -129,7 +142,6 @@ describe("scheduleDeferredModelOverride", () => {
     //#given no message inserted
 
     //#when
-    const { scheduleDeferredModelOverride } = await import("./ultrawork-db-model-override")
     scheduleDeferredModelOverride(
       "msg_nonexistent",
       { providerID: "anthropic", modelID: "claude-opus-4-6" },
@@ -148,7 +160,6 @@ describe("scheduleDeferredModelOverride", () => {
     insertMessage("msg_003", { providerID: "anthropic", modelID: "claude-sonnet-4-6" })
 
     //#when
-    const { scheduleDeferredModelOverride } = await import("./ultrawork-db-model-override")
     scheduleDeferredModelOverride(
       "msg_003",
       { providerID: "anthropic", modelID: "claude-opus-4-6" },
@@ -167,7 +178,6 @@ describe("scheduleDeferredModelOverride", () => {
     getDataDirSpy.mockReturnValue("/nonexistent/path/that/does/not/exist")
 
     //#when
-    const { scheduleDeferredModelOverride } = await import("./ultrawork-db-model-override")
     scheduleDeferredModelOverride(
       "msg_004",
       { providerID: "anthropic", modelID: "claude-opus-4-6" },
@@ -188,7 +198,6 @@ describe("scheduleDeferredModelOverride", () => {
     chmodSync(corruptedDbPath, 0o000)
 
     //#when
-    const { scheduleDeferredModelOverride } = await import("./ultrawork-db-model-override")
     scheduleDeferredModelOverride(
       "msg_corrupt",
       { providerID: "anthropic", modelID: "claude-opus-4-6" },
diff --git a/src/plugin/ultrawork-model-override.test.ts b/src/plugin/ultrawork-model-override.test.ts
index 0204f0cae..feaf369c1 100644
--- a/src/plugin/ultrawork-model-override.test.ts
+++ b/src/plugin/ultrawork-model-override.test.ts
@@ -1,14 +1,29 @@
 import { describe, expect, test, beforeEach, afterEach, spyOn } from "bun:test"
-import {
-  applyUltraworkModelOverrideOnMessage,
-  resolveUltraworkOverride,
-  detectUltrawork,
-} from "./ultrawork-model-override"
 import * as sharedModule from "../shared"
 import * as dbOverrideModule from "./ultrawork-db-model-override"
 import * as sessionStateModule from "../features/claude-code-session-state"
 
+let resolveUltraworkOverride: (typeof import("./ultrawork-model-override"))["resolveUltraworkOverride"]
+let detectUltrawork: (typeof import("./ultrawork-model-override"))["detectUltrawork"]
+let applyUltraworkModelOverrideOnMessage: (typeof import("./ultrawork-model-override"))["applyUltraworkModelOverrideOnMessage"]
+
+async function importFreshUltraworkModelOverrideModule(): Promise<typeof import("./ultrawork-model-override")> {
+  return import(`./ultrawork-model-override?test=${Date.now()}-${Math.random()}`)
+}
+
+async function loadFreshUltraworkModelOverrideModule(): Promise<void> {
+  ;({
+    resolveUltraworkOverride,
+    detectUltrawork,
+    applyUltraworkModelOverrideOnMessage,
+  } = await importFreshUltraworkModelOverrideModule())
+}
+
 describe("detectUltrawork", () => {
+  beforeEach(async () => {
+    await loadFreshUltraworkModelOverrideModule()
+  })
+
   test("should detect ultrawork keyword", () => {
     expect(detectUltrawork("ultrawork do something")).toBe(true)
   })
@@ -41,6 +56,10 @@ describe("detectUltrawork", () => {
 })
 
 describe("resolveUltraworkOverride", () => {
+  beforeEach(async () => {
+    await loadFreshUltraworkModelOverrideModule()
+  })
+
   function createOutput(text: string, agentName?: string) {
     return {
       message: {
@@ -203,7 +222,8 @@ describe("resolveUltraworkOverride", () => {
     //#given
     const config = createConfig("sisyphus", { model: "anthropic/claude-opus-4-6", variant: "max" })
     const output = createOutput("ultrawork do something")
-    const getSessionAgentSpy = spyOn(sessionStateModule, "getSessionAgent").mockReturnValue("sisyphus")
+    const getSessionAgentSpy = spyOn(sessionStateModule, "getSessionAgent")
+    getSessionAgentSpy.mockReturnValue("sisyphus")
 
     //#when
     const result = resolveUltraworkOverride(config, undefined, output, "ses_test")
@@ -220,9 +240,12 @@ describe("applyUltraworkModelOverrideOnMessage", () => {
   let logSpy: ReturnType<typeof spyOn>
   let dbOverrideSpy: ReturnType<typeof spyOn>
 
-  beforeEach(() => {
-    logSpy = spyOn(sharedModule, "log").mockImplementation(() => {})
-    dbOverrideSpy = spyOn(dbOverrideModule, "scheduleDeferredModelOverride").mockImplementation(() => {})
+  beforeEach(async () => {
+    logSpy = spyOn(sharedModule, "log")
+    logSpy.mockImplementation(() => {})
+    dbOverrideSpy = spyOn(dbOverrideModule, "scheduleDeferredModelOverride")
+    dbOverrideSpy.mockImplementation(() => {})
+    await loadFreshUltraworkModelOverrideModule()
   })
 
   afterEach(() => {