@tanzerfe/page-use 0.0.7 → 0.0.8

package/dist/index.mjs

@@ -1,3 +1,3 @@
-import { A as pageUseSettings, B as BROWSER_ENGINE_ID, C as getPageUseConfig, D as pageUsePresets, E as pageUseStore, F as DEFAULT_TOUR_MAX_TAB_SWITCHES, H as DEFAULT_TOUR_STYLES, I as normalizeOptions, L as readLauncherSwitch, M as DEFAULT_MODEL, N as DEFAULT_SESSION_HEADER, O as pageUseHistory, P as DEFAULT_STORAGE_KEY_PREFIX, R as PAGE_USE_MASK_COLORS, S as getElapsedSeconds, T as isRunning, U as TOUR_STYLE_TRAINER, V as browserTtsEngine, W as TOUR_STYLE_USER, _ as runPageUseTool, a as PAGE_USE_STATIC_PRESETS, b as pageUseCapability, c as PAGE_USE_GLOW_TAG, d as PAGE_USE_GLOBAL_KEY, f as PAGE_USE_COMMAND_TYPE, g as isPageUseCommand, h as formatPageUseResultForModel, i as PageUseLauncherElement, j as DEFAULT_MAX_STEPS, k as DEFAULT_PAGE_USE_SETTINGS, l as PageUseGlowElement, m as PAGE_USE_TOOL_NAME, n as mountPageUseUI, o as buildMenuPresets, p as PAGE_USE_TOOL, r as PAGE_USE_LAUNCHER_TAG, s as PAGE_USE_ROBOT_ICON_SVG, t as definePageUseElements, u as createPageUse, v as hidePageUseLauncher, w as isPageUseConfigured, x as clearPageUseLog, y as showPageUseLauncher, z as SIMULATOR_MASK_ID } from "./src-ChJ6q7jl.mjs";
+import { A as pageUseSettings, B as BROWSER_ENGINE_ID, C as getPageUseConfig, D as pageUsePresets, E as pageUseStore, F as DEFAULT_TOUR_MAX_TAB_SWITCHES, H as DEFAULT_TOUR_STYLES, I as normalizeOptions, L as readLauncherSwitch, M as DEFAULT_MODEL, N as DEFAULT_SESSION_HEADER, O as pageUseHistory, P as DEFAULT_STORAGE_KEY_PREFIX, R as PAGE_USE_MASK_COLORS, S as getElapsedSeconds, T as isRunning, U as TOUR_STYLE_TRAINER, V as browserTtsEngine, W as TOUR_STYLE_USER, _ as runPageUseTool, a as PAGE_USE_STATIC_PRESETS, b as pageUseCapability, c as PAGE_USE_GLOW_TAG, d as PAGE_USE_GLOBAL_KEY, f as PAGE_USE_COMMAND_TYPE, g as isPageUseCommand, h as formatPageUseResultForModel, i as PageUseLauncherElement, j as DEFAULT_MAX_STEPS, k as DEFAULT_PAGE_USE_SETTINGS, l as PageUseGlowElement, m as PAGE_USE_TOOL_NAME, n as mountPageUseUI, o as buildMenuPresets, p as PAGE_USE_TOOL, r as PAGE_USE_LAUNCHER_TAG, s as PAGE_USE_ROBOT_ICON_SVG, t as definePageUseElements, u as createPageUse, v as hidePageUseLauncher, w as isPageUseConfigured, x as clearPageUseLog, y as showPageUseLauncher, z as SIMULATOR_MASK_ID } from "./src-DRLJiTQV.mjs";
 
 export { BROWSER_ENGINE_ID, DEFAULT_MAX_STEPS, DEFAULT_MODEL, DEFAULT_PAGE_USE_SETTINGS, DEFAULT_SESSION_HEADER, DEFAULT_STORAGE_KEY_PREFIX, DEFAULT_TOUR_MAX_TAB_SWITCHES, DEFAULT_TOUR_STYLES, PAGE_USE_COMMAND_TYPE, PAGE_USE_GLOBAL_KEY, PAGE_USE_GLOW_TAG, PAGE_USE_LAUNCHER_TAG, PAGE_USE_MASK_COLORS, PAGE_USE_ROBOT_ICON_SVG, PAGE_USE_STATIC_PRESETS, PAGE_USE_TOOL, PAGE_USE_TOOL_NAME, PageUseGlowElement, PageUseLauncherElement, SIMULATOR_MASK_ID, TOUR_STYLE_TRAINER, TOUR_STYLE_USER, browserTtsEngine, buildMenuPresets, clearPageUseLog, createPageUse, definePageUseElements, formatPageUseResultForModel, getElapsedSeconds, getPageUseConfig, hidePageUseLauncher, isPageUseCommand, isPageUseConfigured, isRunning, mountPageUseUI, normalizeOptions, pageUseCapability, pageUseHistory, pageUsePresets, pageUseSettings, pageUseStore, readLauncherSwitch, runPageUseTool, showPageUseLauncher };

package/dist/{src-ChJ6q7jl.mjs → src-DRLJiTQV.mjs}

@@ -1,6 +1,6 @@
 import { PageAgent, tool } from "page-agent";
-import { clickElement } from "@page-agent/page-controller";
 import * as z from "zod/v4";
+import { clickElement } from "@page-agent/page-controller";
 import { LitElement, css, html, nothing, svg } from "lit";
 import { keyed } from "lit/directives/keyed.js";
 import { unsafeHTML } from "lit/directives/unsafe-html.js";
@@ -1596,6 +1596,115 @@ function createPageTour(deps) {
 	};
 }
 
+//#endregion
+//#region src/core/newWindow.ts
+/**
+* 新窗口打开记录器：任务执行期间识别点击触发的新窗口打开，供点击工具与按页提示反馈给模型。
+* - 识别：window.open、点击 <a target>（含点到其子元素）、未插入文档的 <a> 调用 click()
+* - 只以调用发生为依据：不检查 window.open 返回值，被浏览器弹窗拦截也视为已打开
+* - 不改变页面行为：原调用照常执行，参数与返回值不变，不拦截重复打开
+* 与 page-agent 解耦，由 createPageUseAgent 在任务开始时 install、结束时 uninstall。
+*/
+/** 在当前窗口内打开的窗口名（比较时转小写），不视为新窗口 */
+const SAME_WINDOW_TARGETS = /* @__PURE__ */ new Set([
+	"_self",
+	"_parent",
+	"_top"
+]);
+/** 窗口名是否指向新窗口：空串（window.open 默认 _blank）与自定义窗口名均视为新窗口 */
+function opensNewWindow(name) {
+	return !SAME_WINDOW_TARGETS.has(name.trim().toLowerCase());
+}
+/** 链接满足新窗口条件时返回绝对地址，否则返回 null */
+function newWindowAnchorUrl(anchor) {
+	const target = anchor.getAttribute("target") ?? "";
+	if (!target.trim() || !opensNewWindow(target)) return null;
+	if (anchor.hasAttribute("download")) return null;
+	const href = anchor.getAttribute("href")?.trim() ?? "";
+	if (!href || /^javascript:/i.test(href)) return null;
+	return anchor.href || href;
+}
+/** window.open 的地址参数解析为绝对 URL；未传地址时浏览器打开空白页 */
+function resolveOpenUrl(win, url) {
+	const raw = url === void 0 ? "" : String(url).trim();
+	if (!raw) return "about:blank";
+	try {
+		return new URL(raw, win.location.href).href;
+	} catch {
+		return raw;
+	}
+}
+function unique(urls) {
+	return [...new Set(urls)];
+}
+function createNewWindowWatcher(win = window) {
+	let records = [];
+	let installed = null;
+	function recordAnchorClick(anchor, event) {
+		const url = newWindowAnchorUrl(anchor);
+		if (url) records.push({
+			url,
+			event
+		});
+	}
+	/** 已插入文档的链接（含 Shadow DOM 内）：点击事件经 composedPath 传到 window */
+	function onWindowClick(event) {
+		const anchor = event.composedPath().find((node) => node instanceof win.HTMLAnchorElement);
+		if (anchor) recordAnchorClick(anchor, event);
+	}
+	function install() {
+		uninstall();
+		records = [];
+		const originalOpen = win.open;
+		const wrappedOpen = (url, target, features) => {
+			const opened = originalOpen.call(win, url, target, features);
+			if (opensNewWindow(target ?? "")) records.push({ url: resolveOpenUrl(win, url) });
+			return opened;
+		};
+		win.open = wrappedOpen;
+		const proto = win.HTMLAnchorElement.prototype;
+		const originalClick = proto.click;
+		const wrappedClick = function() {
+			if (this.isConnected) return originalClick.call(this);
+			const listener = (event) => recordAnchorClick(this, event);
+			this.addEventListener("click", listener, {
+				capture: true,
+				once: true
+			});
+			try {
+				originalClick.call(this);
+			} finally {
+				this.removeEventListener("click", listener, { capture: true });
+			}
+		};
+		proto.click = wrappedClick;
+		win.addEventListener("click", onWindowClick, { capture: true });
+		installed = { restore: () => {
+			if (win.open === wrappedOpen) win.open = originalOpen;
+			if (proto.click === wrappedClick) proto.click = originalClick;
+			win.removeEventListener("click", onWindowClick, { capture: true });
+		} };
+	}
+	function uninstall() {
+		installed?.restore();
+		installed = null;
+	}
+	function since(n) {
+		return unique(records.slice(n).filter((r) => !r.event?.defaultPrevented).map((r) => r.url));
+	}
+	return {
+		install,
+		uninstall,
+		count: () => records.length,
+		since,
+		all: () => since(0)
+	};
+}
+/** 点击结果追加的说明：告知模型新窗口不可观察，当前页面不会变化 */
+function formatOpenedNotice(urls) {
+	return `已通过新窗口打开：${urls.join("、")}。新窗口不在可操作范围内，当前页面不会因此变化。`;
+}
+
 //#endregion
 //#region src/core/siteMap.ts
 /** 去掉 `#`、前导 `/` 与查询串，得到用于比较的路径 */
@@ -1677,6 +1786,8 @@ function buildSystemInstruction(options, flags = {}) {
 		"- 页面内的 Tab 通过点击 Tab 标签切换；详情/编辑通常是表格行内的按钮，点击后在抽屉或弹窗中打开。",
 		"- 下拉选择、日期选择若为自定义组件而非原生 select：先点击该输入框，再点击弹出层中的目标选项文字。",
 		"- 数据加载需要时间；点击后若内容未变化，先使用 wait 等待 1 到 2 秒再重新观察。",
+		"- 允许点击会在新窗口打开页面或外链的入口（用户任务需要打开它时）。",
+		"- 点击结果或页面提示出现「已通过新窗口打开」时，新窗口内容不可观察，当前页面也不会变化；若打开该页面就是任务目标，立即调用 done（success=true），说明已在新窗口打开的页面或地址，不要重复点击，也不要等待当前页面变化。",
 		"",
 		"## 安全边界",
 		"- 未经用户明确要求，不得执行删除、提交、审批、发布、批量修改等不可逆操作；不确定时用 ask_user 询问。",
@@ -1686,10 +1797,11 @@ function buildSystemInstruction(options, flags = {}) {
 }
 /**
 * 按页提示：page-agent 每步调用一次。
-* 路由名命中 routeHints 优先，其次按路径命中；outlineText（页面讲解的模块清单）追加在末尾。
+* 路由名命中 routeHints 优先，其次按路径命中；outlineText（页面讲解的模块清单）追加在其后；
+* openedUrls（本任务已通过新窗口打开的地址）追加在末尾，覆盖点击结果返回后才发生的异步打开。
 * 返回 null 表示不追加。
 */
-function buildPageInstruction(options, outlineText = "") {
+function buildPageInstruction(options, outlineText = "", openedUrls = []) {
 	const path = options.getCurrentPath();
 	const parts = [];
 	const menu = findMenuByPath(path, options.getMenus());
@@ -1703,6 +1815,7 @@ function buildPageInstruction(options, outlineText = "") {
 	const hint = routeName && options.routeHints[routeName] || options.routeHints[normalizePath(path)];
 	if (hint) parts.push(hint);
 	if (outlineText) parts.push(outlineText);
+	if (openedUrls.length) parts.push(`本任务已通过新窗口打开：${openedUrls.join("、")}`);
 	return parts.length ? parts.join("\n") : null;
 }
 
@@ -1714,11 +1827,30 @@ function buildPageInstruction(options, outlineText = "") {
 * - 站点知识：菜单树写入 system 提示，按页提示按当前路径动态注入
 * - 任务开始前按设置执行宿主的 beforeTask（如展开折叠的侧栏）
 * - 页面讲解：按页提示附带页面模块清单，并注册 present_page 工具（pageTour 关闭时均不启用）
+* - 新窗口打开：覆盖内置点击工具，点击触发新窗口打开时在结果中说明；按页提示列出本任务已打开的地址
 */
 /** DOM 文本超过该长度时截断，避免大表格页把上下文撑爆 */
 const PAGE_CONTENT_MAX_CHARS = 6e4;
 /** 每步间隔（秒）：留给路由切换与接口加载的时间 */
 const STEP_DELAY_SECONDS = .6;
+/** page-agent 内置点击工具名：同名注册即覆盖内置实现 */
+const CLICK_TOOL_NAME = "click_element_by_index";
+/**
+* 点击工具：与内置实现相同（描述、输入结构不变），点击期间识别到新窗口打开时在结果后追加说明。
+* clickElement 返回前已等待 0.2 秒，同步与微任务内的打开均已记录；更晚的打开由按页提示兜底。
+*/
+function createClickTool(watcher) {
+	return tool({
+		description: "Click element by index",
+		inputSchema: z.object({ index: z.int().min(0) }),
+		async execute(input) {
+			const before = watcher.count();
+			const { message } = await this.pageController.clickElement(input.index);
+			const urls = watcher.since(before);
+			return urls.length ? `${message}\n${formatOpenedNotice(urls)}` : message;
+		}
+	});
+}
 /** 生成会话 ID；一次 execute 视为一个会话，任务开始时刷新 */
 function newSessionId() {
 	return typeof crypto !== "undefined" && crypto.randomUUID?.() || `page-use-${Date.now()}`;
@@ -1749,6 +1881,7 @@ function createPageUseAgent({ options, enableMask, narration = false, shouldRunB
 		onMaskForced,
 		getStyle: getTourStyle
 	}) : null;
+	const newWindows = createNewWindowWatcher();
 	return new PageAgent({
 		baseURL: llm.baseURL,
 		apiKey: llm.apiKey || void 0,
@@ -1767,7 +1900,7 @@ function createPageUseAgent({ options, enableMask, narration = false, shouldRunB
 		experimentalScriptExecutionTool: false,
 		instructions: {
 			system: buildSystemInstruction(options, { narration }),
-			getPageInstructions: () => buildPageInstruction(options, tour?.outlineText() ?? "")
+			getPageInstructions: () => buildPageInstruction(options, tour?.outlineText() ?? "", newWindows.all())
 		},
 		transformPageContent: (content) => {
 			if (content.length <= PAGE_CONTENT_MAX_CHARS) return content;
@@ -1775,10 +1908,12 @@ function createPageUseAgent({ options, enableMask, narration = false, shouldRunB
 		},
 		...hostPageAgent,
 		customTools: {
+			[CLICK_TOOL_NAME]: createClickTool(newWindows),
 			...tour ? { [PRESENT_PAGE_TOOL_NAME]: tour.tool } : {},
 			...hostPageAgent.customTools
 		},
 		onBeforeTask: async (...args) => {
+			newWindows.install();
 			sessionId = newSessionId();
 			tour?.onBeforeTask();
 			if (hostBeforeTask && shouldRunBeforeTask()) try {
@@ -1789,9 +1924,14 @@ function createPageUseAgent({ options, enableMask, narration = false, shouldRunB
 			await hostPageAgent.onBeforeTask?.(...args);
 		},
 		onAfterTask: async (...args) => {
+			newWindows.uninstall();
 			await tour?.onAfterTask(args[1]);
 			onTaskFinished?.(args[1]);
 			await hostPageAgent.onAfterTask?.(...args);
+		},
+		onDispose: (...args) => {
+			newWindows.uninstall();
+			hostPageAgent.onDispose?.(...args);
 		}
 	});
 }
@@ -2904,7 +3044,7 @@ var PageUseGlowElement = class extends LitElement {
 
 //#endregion
 //#region package.json
-var version = "0.0.7";
+var version = "0.0.8";
 
 //#endregion
 //#region src/ui/drag.ts

package/dist/vue.mjs

@@ -1,4 +1,4 @@
-import { A as pageUseSettings, B as BROWSER_ENGINE_ID, C as getPageUseConfig, D as pageUsePresets, E as pageUseStore, F as DEFAULT_TOUR_MAX_TAB_SWITCHES, H as DEFAULT_TOUR_STYLES, I as normalizeOptions, L as readLauncherSwitch, M as DEFAULT_MODEL, N as DEFAULT_SESSION_HEADER, O as pageUseHistory, P as DEFAULT_STORAGE_KEY_PREFIX, R as PAGE_USE_MASK_COLORS, S as getElapsedSeconds, T as isRunning, U as TOUR_STYLE_TRAINER, V as browserTtsEngine, W as TOUR_STYLE_USER, _ as runPageUseTool, a as PAGE_USE_STATIC_PRESETS, b as pageUseCapability, c as PAGE_USE_GLOW_TAG, d as PAGE_USE_GLOBAL_KEY, f as PAGE_USE_COMMAND_TYPE, g as isPageUseCommand, h as formatPageUseResultForModel, i as PageUseLauncherElement, j as DEFAULT_MAX_STEPS, k as DEFAULT_PAGE_USE_SETTINGS, l as PageUseGlowElement, m as PAGE_USE_TOOL_NAME, n as mountPageUseUI, o as buildMenuPresets, p as PAGE_USE_TOOL, r as PAGE_USE_LAUNCHER_TAG, s as PAGE_USE_ROBOT_ICON_SVG, t as definePageUseElements, u as createPageUse, v as hidePageUseLauncher, w as isPageUseConfigured, x as clearPageUseLog, y as showPageUseLauncher, z as SIMULATOR_MASK_ID } from "./src-ChJ6q7jl.mjs";
+import { A as pageUseSettings, B as BROWSER_ENGINE_ID, C as getPageUseConfig, D as pageUsePresets, E as pageUseStore, F as DEFAULT_TOUR_MAX_TAB_SWITCHES, H as DEFAULT_TOUR_STYLES, I as normalizeOptions, L as readLauncherSwitch, M as DEFAULT_MODEL, N as DEFAULT_SESSION_HEADER, O as pageUseHistory, P as DEFAULT_STORAGE_KEY_PREFIX, R as PAGE_USE_MASK_COLORS, S as getElapsedSeconds, T as isRunning, U as TOUR_STYLE_TRAINER, V as browserTtsEngine, W as TOUR_STYLE_USER, _ as runPageUseTool, a as PAGE_USE_STATIC_PRESETS, b as pageUseCapability, c as PAGE_USE_GLOW_TAG, d as PAGE_USE_GLOBAL_KEY, f as PAGE_USE_COMMAND_TYPE, g as isPageUseCommand, h as formatPageUseResultForModel, i as PageUseLauncherElement, j as DEFAULT_MAX_STEPS, k as DEFAULT_PAGE_USE_SETTINGS, l as PageUseGlowElement, m as PAGE_USE_TOOL_NAME, n as mountPageUseUI, o as buildMenuPresets, p as PAGE_USE_TOOL, r as PAGE_USE_LAUNCHER_TAG, s as PAGE_USE_ROBOT_ICON_SVG, t as definePageUseElements, u as createPageUse, v as hidePageUseLauncher, w as isPageUseConfigured, x as clearPageUseLog, y as showPageUseLauncher, z as SIMULATOR_MASK_ID } from "./src-DRLJiTQV.mjs";
 import { defineComponent, inject, onBeforeUnmount, onScopeDispose, shallowRef } from "vue";
 import { routerKey } from "vue-router";
 

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@tanzerfe/page-use",
   "private": false,
-  "version": "0.0.7",
+  "version": "0.0.8",
   "description": "基于 page-agent 的页面操作智能体：自然语言指令驱动页面点击与输入，框架无关，附 Web Component 调试面板与 Vue 适配层",
   "type": "module",
   "main": "./dist/index.mjs",