dsh-browser-playwright-codex 0.5.2 → 0.5.3

package/lib/url-policy.js

@@ -0,0 +1,26 @@
+/**
+ * URL 策略：只允许访问白名单域名（含点击后导航的校验）。
+ * @module dsh-browser-playwright-codex/url-policy
+ */
+import { BrowserError } from "./errors.js";
+/** Validate one absolute URL against the navigation policy. */
+export function assertAllowedUrl(raw, allowedDomains) {
+    let parsed;
+    try {
+        parsed = new URL(raw);
+    }
+    catch {
+        throw new BrowserError('URL_NOT_ALLOWED', 'invalid URL: ' + raw);
+    }
+    if (parsed.protocol !== 'http:' && parsed.protocol !== 'https:') {
+        throw new BrowserError('URL_NOT_ALLOWED', 'only http(s) URLs can be navigated, got: ' + parsed.protocol);
+    }
+    if (allowedDomains.length > 0) {
+        const host = parsed.hostname.toLowerCase();
+        const allowed = allowedDomains.some((suffix) => host === suffix || host.endsWith('.' + suffix));
+        if (!allowed) {
+            throw new BrowserError('URL_NOT_ALLOWED', 'host ' + parsed.hostname + ' is not in allowedDomains');
+        }
+    }
+    return parsed;
+}

package/lib/window-focus.d.ts

@@ -0,0 +1,27 @@
+/**
+ * 窗口与焦点行为：判断承载页面的系统窗口是否最小化，以及在后台创建标签页 —— 两项都是为了让
+ * 窗口/标签不打扰用户。背景与实测见 FOCUS-STEALING.md。
+ * @module dsh-browser-playwright-codex/window-focus
+ */
+import type { BrowserContext, Page } from 'playwright-core';
+/**
+ * True when the OS window hosting this page is minimized. Tab selection lives
+ * in the browser context, not in the OS window, so activation is a side effect
+ * we are free to decline — and declining it is what keeps the agent from
+ * yanking a minimized window into the user's face. If the window state cannot
+ * be read we answer true (treat as minimized): the safe direction is to leave
+ * the user's desktop alone, and a skipped bringToFront costs nothing functional.
+ */
+export declare function isWindowMinimized(page: Page): Promise<boolean>;
+/**
+ * Create a tab without raising the window. context.newPage() routes through CDP
+ * Target.createTarget without `background`, and Chromium activates the window
+ * on every tab creation -- plainly visible once the user has minimized it.
+ * Sending the command ourselves with background:true creates the tab silently;
+ * Playwright still auto-attaches and returns a fully driveable Page (verified
+ * live: Playwright learns the page and evaluate() runs on it). The URL is left
+ * at about:blank on purpose so the caller's own goto keeps owning navigation
+ * policy, timeouts and abort handling. On any failure this falls back to
+ * newPage(): a tab that pops the window beats no tab at all.
+ */
+export declare function createBackgroundPage(context: BrowserContext): Promise<Page>;

package/lib/window-focus.js

@@ -0,0 +1,54 @@
+/**
+ * True when the OS window hosting this page is minimized. Tab selection lives
+ * in the browser context, not in the OS window, so activation is a side effect
+ * we are free to decline — and declining it is what keeps the agent from
+ * yanking a minimized window into the user's face. If the window state cannot
+ * be read we answer true (treat as minimized): the safe direction is to leave
+ * the user's desktop alone, and a skipped bringToFront costs nothing functional.
+ */
+export async function isWindowMinimized(page) {
+    let provider;
+    try {
+        provider = await page.context().newCDPSession(page);
+        const { windowId } = await provider.send('Browser.getWindowForTarget');
+        const { bounds } = await provider.send('Browser.getWindowBounds', { windowId });
+        return bounds.windowState === 'minimized';
+    }
+    catch {
+        return true;
+    }
+    finally {
+        await provider?.detach().catch(() => { });
+    }
+}
+/**
+ * Create a tab without raising the window. context.newPage() routes through CDP
+ * Target.createTarget without `background`, and Chromium activates the window
+ * on every tab creation -- plainly visible once the user has minimized it.
+ * Sending the command ourselves with background:true creates the tab silently;
+ * Playwright still auto-attaches and returns a fully driveable Page (verified
+ * live: Playwright learns the page and evaluate() runs on it). The URL is left
+ * at about:blank on purpose so the caller's own goto keeps owning navigation
+ * policy, timeouts and abort handling. On any failure this falls back to
+ * newPage(): a tab that pops the window beats no tab at all.
+ */
+export async function createBackgroundPage(context) {
+    const anchor = context.pages()[0];
+    if (anchor === undefined)
+        return await context.newPage();
+    let provider;
+    try {
+        provider = await context.newCDPSession(anchor);
+        // Arm the listener before sending: Playwright attaches as soon as the
+        // target exists, which can happen before createTarget resolves.
+        const attached = context.waitForEvent('page', { timeout: 10000 });
+        await provider.send('Target.createTarget', { url: 'about:blank', background: true });
+        return await attached;
+    }
+    catch {
+        return await context.newPage();
+    }
+    finally {
+        await provider?.detach().catch(() => { });
+    }
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-browser-playwright-codex",
-  "version": "0.5.2",
+  "version": "0.5.3",
   "description": "Browser automation for DeepSeek Harness: fill in forms, scrape lists and work across tabs in real pages, using the logins you already have. Runs in a visible window that stays quiet when minimised, and stops for sign-ins, CAPTCHAs and native dialogs instead of confirming for you.",
   "license": "MIT",
   "type": "module",

package/scripts/dsh-browser-verify.mjs

@@ -23,7 +23,8 @@ import http from 'node:http'
 import { readdirSync, statSync } from 'node:fs'
 import os from 'node:os'
 import path from 'node:path'
-import { Config as ConfigSchema, PlaywrightProvider } from '../lib/playwright.js'
+import { PlaywrightProvider } from '../lib/provider.js'
+import { Config as ConfigSchema } from '../lib/config.js'
 import { renderSnapshot } from '../lib/snapshot-render.js'
 import { getEffectiveEngine, setSnapshotEngine } from '../lib/runtime-state.js'
 

package/scripts/prompt-cost.mjs

@@ -12,7 +12,7 @@ import { Context } from '@deepseek-ai/cordis'
 import ToolRuntime from '@deepseek-ai/dsh-tools'
 import SystemPrompt from '@deepseek-ai/dsh-system-prompt'
 import BrowserRuntime from '../src/service.ts'
-import { PlaywrightProvider } from '../src/playwright.ts'
+import { PlaywrightProvider } from '../src/provider.ts'
 import * as browserTool from '../src/tool.ts'
 import { bundleConfig } from './shipped-gates.mjs'