feat(hooks): add unstable-agent-babysitter hook for monitoring unstable background agents

2026-02-01 16:59:50 +09:00
parent 9b60f08f82
commit a0e97ce5c0
5 changed files with 425 additions and 0 deletions
@@ -0,0 +1,142 @@
+import { _resetForTesting, setMainSession } from "../../features/claude-code-session-state"
+import type { BackgroundTask } from "../../features/background-agent"
+import { createUnstableAgentBabysitterHook } from "./index"
+
+const projectDir = "/Users/yeongyu/local-workspaces/oh-my-opencode"
+
+type BabysitterContext = Parameters<typeof createUnstableAgentBabysitterHook>[0]
+
+function createMockPluginInput(options: {
+  messagesBySession: Record<string, unknown[]>
+  promptCalls: Array<{ input: unknown }>
+}): BabysitterContext {
+  const { messagesBySession, promptCalls } = options
+  return {
+    directory: projectDir,
+    client: {
+      session: {
+        messages: async ({ path }: { path: { id: string } }) => ({
+          data: messagesBySession[path.id] ?? [],
+        }),
+        prompt: async (input: unknown) => {
+          promptCalls.push({ input })
+        },
+      },
+    },
+  }
+}
+
+function createBackgroundManager(tasks: BackgroundTask[]) {
+  return {
+    getTasksByParentSession: () => tasks,
+  }
+}
+
+function createTask(overrides: Partial<BackgroundTask> = {}): BackgroundTask {
+  return {
+    id: "task-1",
+    sessionID: "bg-1",
+    parentSessionID: "main-1",
+    parentMessageID: "msg-1",
+    description: "unstable task",
+    prompt: "run work",
+    agent: "test-agent",
+    status: "running",
+    progress: {
+      toolCalls: 1,
+      lastUpdate: new Date(),
+      lastMessage: "still working",
+      lastMessageAt: new Date(Date.now() - 121000),
+    },
+    model: { providerID: "google", modelID: "gemini-1.5" },
+    ...overrides,
+  }
+}
+
+describe("unstable-agent-babysitter hook", () => {
+  afterEach(() => {
+    _resetForTesting()
+  })
+
+  test("fires reminder for hung gemini task", async () => {
+    // #given
+    setMainSession("main-1")
+    const promptCalls: Array<{ input: unknown }> = []
+    const ctx = createMockPluginInput({
+      messagesBySession: {
+        "main-1": [
+          { info: { agent: "sisyphus", model: { providerID: "openai", modelID: "gpt-4" } } },
+        ],
+        "bg-1": [
+          { info: { role: "assistant" }, parts: [{ type: "thinking", thinking: "deep thought" }] },
+        ],
+      },
+      promptCalls,
+    })
+    const backgroundManager = createBackgroundManager([createTask()])
+    const hook = createUnstableAgentBabysitterHook(ctx, {
+      backgroundManager,
+      config: { enabled: true, timeout_ms: 120000 },
+    })
+
+    // #when
+    await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
+
+    // #then
+    expect(promptCalls.length).toBe(1)
+    const payload = promptCalls[0].input as { body?: { parts?: Array<{ text?: string }> } }
+    const text = payload.body?.parts?.[0]?.text ?? ""
+    expect(text).toContain("background_output")
+    expect(text).toContain("background_cancel")
+    expect(text).toContain("deep thought")
+  })
+
+  test("does not remind stable model tasks", async () => {
+    // #given
+    setMainSession("main-1")
+    const promptCalls: Array<{ input: unknown }> = []
+    const ctx = createMockPluginInput({
+      messagesBySession: { "main-1": [] },
+      promptCalls,
+    })
+    const backgroundManager = createBackgroundManager([
+      createTask({ model: { providerID: "openai", modelID: "gpt-4" } }),
+    ])
+    const hook = createUnstableAgentBabysitterHook(ctx, {
+      backgroundManager,
+      config: { enabled: true, timeout_ms: 120000 },
+    })
+
+    // #when
+    await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
+
+    // #then
+    expect(promptCalls.length).toBe(0)
+  })
+
+  test("respects per-task cooldown", async () => {
+    // #given
+    setMainSession("main-1")
+    const promptCalls: Array<{ input: unknown }> = []
+    const ctx = createMockPluginInput({
+      messagesBySession: { "main-1": [], "bg-1": [] },
+      promptCalls,
+    })
+    const backgroundManager = createBackgroundManager([createTask()])
+    const hook = createUnstableAgentBabysitterHook(ctx, {
+      backgroundManager,
+      config: { enabled: true, timeout_ms: 120000 },
+    })
+    const now = Date.now()
+    const originalNow = Date.now
+    Date.now = () => now
+
+    // #when
+    await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
+    await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
+
+    // #then
+    expect(promptCalls.length).toBe(1)
+    Date.now = originalNow
+  })
+})