✨ 新增 get_task/delete_task 工具，优化 Agent 工具提示词分层

- task_tools: 新增 get_task（获取完整详情）和 delete_task（删除任务）
- 工具 description 精简为能力描述，行为指导移入系统提示词
- 系统提示词 Task Management 重构为 When/When NOT/Workflow/Tips 结构
- ask_user description 精简，推荐选项规范移入系统提示词
- 新增 6 个 task_tools 测试用例

Author: CodFrm

src/app/service/agent/system_prompt.ts

@@ -63,7 +63,7 @@ When stuck, **prioritize asking the user over repeated attempts**:
 - **Interact with page DOM** → \`execute_script(target='page')\` for clicking, filling forms, reading dynamic state. Runs in MAIN world (shares page globals). Use \`get_tab_content\` first to understand page structure.
 - **Compute without DOM** → \`execute_script(target='sandbox')\` for data processing, text parsing, calculations.
 - **Search the web** → \`web_search\` returns titles, URLs, and snippets. Follow up with \`web_fetch\` to read specific results.
-- **Ask user** → \`ask_user\` for questions. Prefer providing \`options\` for structured choices so the user can select quickly; add \`multiple: true\` for multi-select. The user can also type a custom response even when options are provided.
+- **Ask user** → \`ask_user\` to gather preferences, clarify ambiguous instructions, or get decisions on implementation choices. Prefer providing \`options\` for structured choices so the user can select quickly; add \`multiple: true\` for multi-select. If you recommend a specific option, put it first and append "(Recommended)". The user can always type a custom response even when options are provided.
 
 ## Sub-Agent
 
@@ -83,13 +83,26 @@ Use the \`agent\` tool to delegate **independent subtasks** that don't require u
 
 ## Task Management
 
-For **complex, multi-step tasks**, use task tools to track your progress:
-- \`create_task\` — Break the work into individual steps at the start.
-- \`update_task\` — Mark each step as \`in_progress\` when you begin it, and \`completed\` when done.
-- \`list_tasks\` — Review remaining steps, especially after resuming a conversation.
+Use task tools to create a structured task list that tracks your progress. This helps the user understand what you're doing and how much work remains.
 
-**When to use:** Tasks that involve 3+ distinct steps (e.g., navigating multiple pages, processing data, multi-stage workflows). Do NOT create tasks for simple, single-step requests.
-**Workflow:** Create all tasks first → work through them one by one → update status as you go.
+**When to use:**
+- Complex tasks requiring 3+ distinct steps (e.g., navigating multiple pages, multi-stage data processing)
+- The user provides multiple things to do at once
+- After receiving new instructions — immediately capture requirements as tasks
+
+**When NOT to use:**
+- Single, straightforward tasks that complete in 1-2 steps
+- Purely conversational or informational requests
+
+**Workflow:**
+1. **Plan** — Call \`list_tasks\` to check for existing tasks, then \`create_task\` for each step with a clear imperative subject and enough description for context.
+2. **Execute** — Before starting each task, call \`update_task\` with \`status: "in_progress"\`. When done, set \`status: "completed"\`.
+3. **Adapt** — If a completed task reveals follow-up work, create new tasks. If a task becomes irrelevant, use \`delete_task\` to clean up. Use \`get_task\` to review a task's full description before starting it.
+
+**Tips:**
+- Write subjects as brief imperatives: "Extract product prices", not "I will extract prices".
+- Include acceptance criteria in the description so progress is unambiguous.
+- Do not create tasks you intend to complete in the same tool call — tasks are for tracking multi-step progress, not logging what you already did.
 
 ## OPFS Workspace
 

src/app/service/agent/tools/ask_user.ts

@@ -4,21 +4,21 @@ import type { ToolExecutor } from "@App/app/service/agent/tool_registry";
 export const ASK_USER_DEFINITION: ToolDefinition = {
   name: "ask_user",
   description:
-    "Ask the user a question and wait for their response (text only, no image support). " +
-    "Use options for structured choices (single/multi-select). Times out after 5 minutes.",
+    "Ask the user a question and wait for their response. " +
+    "Text response only (no image support). Times out after 5 minutes. " +
+    "The user can always type a custom response even when options are provided.",
   parameters: {
     type: "object",
     properties: {
       question: { type: "string", description: "The question to ask the user" },
       options: {
         type: "array",
         items: { type: "string" },
-        description:
-          "Optional list of choices for the user. If provided, user selects from these instead of free text input.",
+        description: "List of choices. User selects from these but can also type a custom response.",
       },
       multiple: {
         type: "boolean",
-        description: "Allow selecting multiple options (default: false, single-select).",
+        description: "Allow selecting multiple options (default: false).",
       },
     },
     required: ["question"],

src/app/service/agent/tools/task_tools.test.ts

@@ -2,11 +2,11 @@ import { describe, it, expect, vi } from "vitest";
 import { createTaskTools, type Task } from "./task_tools";
 
 describe("task_tools", () => {
-  it("应创建 3 个工具", () => {
+  it("应创建 5 个工具", () => {
     const { tools } = createTaskTools();
-    expect(tools).toHaveLength(3);
+    expect(tools).toHaveLength(5);
     const names = tools.map((t) => t.definition.name);
-    expect(names).toEqual(["create_task", "update_task", "list_tasks"]);
+    expect(names).toEqual(["create_task", "update_task", "get_task", "delete_task", "list_tasks"]);
   });
 
   it("create_task 应创建自增 ID 的任务", async () => {
@@ -132,6 +132,81 @@ describe("task_tools", () => {
     expect(sendEvent).not.toHaveBeenCalled();
   });
 
+  it("get_task 应返回任务完整信息", async () => {
+    const { tools } = createTaskTools();
+    const create = tools.find((t) => t.definition.name === "create_task")!;
+    const get = tools.find((t) => t.definition.name === "get_task")!;
+
+    await create.executor.execute({ subject: "Task", description: "Detailed info" });
+
+    const result = JSON.parse((await get.executor.execute({ task_id: "1" })) as string);
+    expect(result).toEqual({ id: "1", subject: "Task", description: "Detailed info", status: "pending" });
+  });
+
+  it("get_task 应对不存在的任务抛错", async () => {
+    const { tools } = createTaskTools();
+    const get = tools.find((t) => t.definition.name === "get_task")!;
+    await expect(get.executor.execute({ task_id: "999" })).rejects.toThrow('Task "999" not found');
+  });
+
+  it("delete_task 应删除任务", async () => {
+    const { tools } = createTaskTools();
+    const create = tools.find((t) => t.definition.name === "create_task")!;
+    const del = tools.find((t) => t.definition.name === "delete_task")!;
+    const list = tools.find((t) => t.definition.name === "list_tasks")!;
+
+    await create.executor.execute({ subject: "A" });
+    await create.executor.execute({ subject: "B" });
+
+    const result = JSON.parse((await del.executor.execute({ task_id: "1" })) as string);
+    expect(result).toEqual({ deleted: true, task_id: "1" });
+
+    const remaining = JSON.parse((await list.executor.execute({})) as string);
+    expect(remaining).toHaveLength(1);
+    expect(remaining[0].id).toBe("2");
+  });
+
+  it("delete_task 应对不存在的任务抛错", async () => {
+    const { tools } = createTaskTools();
+    const del = tools.find((t) => t.definition.name === "delete_task")!;
+    await expect(del.executor.execute({ task_id: "999" })).rejects.toThrow('Task "999" not found');
+  });
+
+  it("delete_task 应调用 onSave 和 sendEvent", async () => {
+    const onSave = vi.fn().mockResolvedValue(undefined);
+    const sendEvent = vi.fn();
+    const { tools } = createTaskTools({ onSave, sendEvent });
+    const create = tools.find((t) => t.definition.name === "create_task")!;
+    const del = tools.find((t) => t.definition.name === "delete_task")!;
+
+    await create.executor.execute({ subject: "Task" });
+    onSave.mockClear();
+    sendEvent.mockClear();
+
+    await del.executor.execute({ task_id: "1" });
+
+    expect(onSave).toHaveBeenCalledOnce();
+    expect(onSave).toHaveBeenCalledWith([]);
+    expect(sendEvent).toHaveBeenCalledOnce();
+    expect(sendEvent).toHaveBeenCalledWith({
+      type: "task_update",
+      tasks: [],
+    });
+  });
+
+  it("get_task 不应触发 onSave 或 sendEvent", async () => {
+    const onSave = vi.fn().mockResolvedValue(undefined);
+    const sendEvent = vi.fn();
+    const initial: Task[] = [{ id: "1", subject: "Existing", status: "pending", description: "Desc" }];
+    const { tools } = createTaskTools({ initialTasks: initial, onSave, sendEvent });
+    const get = tools.find((t) => t.definition.name === "get_task")!;
+
+    await get.executor.execute({ task_id: "1" });
+
+    expect(onSave).not.toHaveBeenCalled();
+    expect(sendEvent).not.toHaveBeenCalled();
+  });
+
   it("多实例应独立", async () => {
     const instance1 = createTaskTools();
     const instance2 = createTaskTools();

src/app/service/agent/tools/task_tools.ts

@@ -10,22 +10,26 @@ export type Task = {
 
 const CREATE_TASK_DEFINITION: ToolDefinition = {
   name: "create_task",
-  description:
-    "Create a new task to track work progress. Use this to break complex, multi-step work into trackable steps. Returns the created task with an auto-assigned ID.",
+  description: "Create a new task. Returns the created task with an auto-assigned ID and status 'pending'.",
   parameters: {
     type: "object",
     properties: {
-      subject: { type: "string", description: "Brief title for the task" },
-      description: { type: "string", description: "Detailed description of what needs to be done" },
+      subject: {
+        type: "string",
+        description: "Brief, actionable title in imperative form (e.g., 'Extract product prices from page')",
+      },
+      description: {
+        type: "string",
+        description: "Detailed description including context and acceptance criteria",
+      },
     },
     required: ["subject"],
   },
 };
 
 const UPDATE_TASK_DEFINITION: ToolDefinition = {
   name: "update_task",
-  description:
-    'Update a task\'s status or details. Set status to "in_progress" when starting work, "completed" when done.',
+  description: "Update a task's status or details. Can change status, subject, and description.",
   parameters: {
     type: "object",
     properties: {
@@ -42,9 +46,33 @@ const UPDATE_TASK_DEFINITION: ToolDefinition = {
   },
 };
 
+const GET_TASK_DEFINITION: ToolDefinition = {
+  name: "get_task",
+  description: "Get a task's full details including description. list_tasks only returns id/subject/status.",
+  parameters: {
+    type: "object",
+    properties: {
+      task_id: { type: "string", description: "The task ID" },
+    },
+    required: ["task_id"],
+  },
+};
+
+const DELETE_TASK_DEFINITION: ToolDefinition = {
+  name: "delete_task",
+  description: "Delete a task permanently.",
+  parameters: {
+    type: "object",
+    properties: {
+      task_id: { type: "string", description: "The task ID to delete" },
+    },
+    required: ["task_id"],
+  },
+};
+
 const LIST_TASKS_DEFINITION: ToolDefinition = {
   name: "list_tasks",
-  description: "List all tasks with their IDs, subjects, and statuses. Use to review remaining work.",
+  description: "List all tasks with their IDs, subjects, and statuses (without descriptions).",
   parameters: {
     type: "object",
     properties: {},
@@ -125,6 +153,28 @@ export function createTaskTools(options?: TaskToolsOptions): {
     },
   };
 
+  const getExecutor: ToolExecutor = {
+    execute: async (args: Record<string, unknown>) => {
+      const task = tasks.get(args.task_id as string);
+      if (!task) {
+        throw new Error(`Task "${args.task_id}" not found`);
+      }
+      return JSON.stringify(task);
+    },
+  };
+
+  const deleteExecutor: ToolExecutor = {
+    execute: async (args: Record<string, unknown>) => {
+      const taskId = args.task_id as string;
+      if (!tasks.has(taskId)) {
+        throw new Error(`Task "${taskId}" not found`);
+      }
+      tasks.delete(taskId);
+      await emitUpdate();
+      return JSON.stringify({ deleted: true, task_id: taskId });
+    },
+  };
+
   const listExecutor: ToolExecutor = {
     execute: async () => {
       const list = Array.from(tasks.values()).map((t) => ({
@@ -140,6 +190,8 @@ export function createTaskTools(options?: TaskToolsOptions): {
     tools: [
       { definition: CREATE_TASK_DEFINITION, executor: createExecutor },
       { definition: UPDATE_TASK_DEFINITION, executor: updateExecutor },
+      { definition: GET_TASK_DEFINITION, executor: getExecutor },
+      { definition: DELETE_TASK_DEFINITION, executor: deleteExecutor },
       { definition: LIST_TASKS_DEFINITION, executor: listExecutor },
     ],
     tasks,