@mind-harness/mh-browser 0.2.2 → 0.2.3

package/extension/manifest.json

@@ -1,7 +1,7 @@
 {
   "manifest_version": 3,
   "name": "Mind Harness",
-  "version": "0.2.2",
+  "version": "0.2.3",
   "permissions": ["tabGroups"],
   "background": { "service_worker": "sw.js" }
 }

package/extension/welcome.css

@@ -0,0 +1,94 @@
+/* Keep in step with the sign-in callback page in packages/mh-cloud/src/page.ts. */
+:root {
+  --bg: #f4f4f6;
+  --card: #fff;
+  --stroke: rgba(0, 0, 0, 0.08);
+  --text: #18181b;
+  --muted: #6b6b73;
+  --accent: #4d6bfe;
+}
+
+@media (prefers-color-scheme: dark) {
+  :root {
+    --bg: #161618;
+    --card: #202023;
+    --stroke: rgba(255, 255, 255, 0.08);
+    --text: #ededf0;
+    --muted: #9a9aa3;
+    --accent: #7f95ff;
+  }
+}
+
+* {
+  box-sizing: border-box;
+}
+
+html,
+body {
+  height: 100%;
+}
+
+body {
+  margin: 0;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  padding: 24px;
+  background: var(--bg);
+  color: var(--text);
+  font:
+    15px/1.6 -apple-system,
+    BlinkMacSystemFont,
+    'Segoe UI',
+    'PingFang SC',
+    'Hiragino Sans GB',
+    'Microsoft YaHei',
+    sans-serif;
+}
+
+.card {
+  width: 100%;
+  max-width: 420px;
+  padding: 36px 32px 28px;
+  border: 1px solid var(--stroke);
+  border-radius: 20px;
+  background: var(--card);
+  text-align: center;
+}
+
+.product {
+  margin: 0 0 24px;
+  font-size: 12px;
+  letter-spacing: 0.06em;
+  text-transform: uppercase;
+  color: var(--muted);
+}
+
+.mark {
+  width: 56px;
+  height: 56px;
+  margin: 0 auto 16px;
+  border-radius: 50%;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  color: var(--accent);
+  background: color-mix(in srgb, var(--accent) 14%, transparent);
+}
+
+h1 {
+  margin: 0 0 8px;
+  font-size: 20px;
+  font-weight: 600;
+}
+
+.line {
+  margin: 0 0 4px;
+  color: var(--muted);
+}
+
+.hint {
+  margin: 12px 0 0;
+  font-size: 13px;
+  color: var(--muted);
+}

package/extension/welcome.html

@@ -0,0 +1,40 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <meta name="color-scheme" content="light dark" />
+    <title data-text="tab">Mind Harness browser</title>
+    <link rel="stylesheet" href="welcome.css" />
+    <script src="welcome.js" defer></script>
+  </head>
+  <body>
+    <main class="card">
+      <p class="product">Mind Harness</p>
+      <div class="mark" aria-hidden="true">
+        <svg
+          viewBox="0 0 24 24"
+          width="28"
+          height="28"
+          fill="none"
+          stroke="currentColor"
+          stroke-width="2.2"
+          stroke-linecap="round"
+          stroke-linejoin="round"
+        >
+          <rect x="3.5" y="5" width="17" height="14" rx="2.5" />
+          <path d="M3.5 9.5h17" />
+        </svg>
+      </div>
+      <h1 data-text="title">This is the Mind&nbsp;Harness browser</h1>
+      <p class="line" data-text="what">
+        The agent opens pages, reads them, and acts on them here. Sites you sign in to here stay
+        signed in for the agent.
+      </p>
+      <p class="line" data-text="keep">
+        Please keep this browser running. When the agent needs you, the conversation will tell you.
+      </p>
+      <p class="hint" data-text="closes">This window closes when the agent opens its first page.</p>
+    </main>
+  </body>
+</html>

package/extension/welcome.js

@@ -0,0 +1,15 @@
+const ZH = {
+  tab: 'Mind Harness 浏览器',
+  title: '这是 Mind\u00a0Harness 的浏览器',
+  what: 'Agent 在这里替你打开网页、读取内容、完成操作。你在这里登录过的网站，Agent 之后也能用上。',
+  keep: '请不要退出这个浏览器。需要你来接手时，对话里会提示你。',
+  closes: 'Agent 打开第一个网页后，这个窗口会自动关闭。',
+}
+
+if (navigator.language.toLowerCase().startsWith('zh')) {
+  document.documentElement.lang = 'zh-CN'
+  for (const node of document.querySelectorAll('[data-text]')) {
+    const text = ZH[node.getAttribute('data-text')]
+    if (text !== undefined) node.textContent = text
+  }
+}

package/lib/index.js

@@ -7,7 +7,7 @@ import { assertSupportedJsonSchema, defineTool, validateJsonSchemaValue } from "
 import { createHash, randomUUID } from "node:crypto";
 import { parse } from "yaml";
 import { constants, readFileSync } from "node:fs";
-import { lstat, mkdir, open, readFile, realpath, unlink, writeFile } from "node:fs/promises";
+import { lstat, mkdir, open, readFile, readlink, realpath, unlink, writeFile } from "node:fs/promises";
 import { z as z$1 } from "zod";
 import { Remote, TypertRemoteService } from "@deepseek-ai/dsh-typert-protocol";
 import { chromium } from "playwright";
@@ -2979,6 +2979,50 @@ async function launchBrowser(options) {
 		throw error;
 	}
 }
+/**
+* Process id of the browser's main process. Playwright gives none for a
+* persistent context. Asks the browser first; on POSIX the profile's
+* `SingletonLock` link names it as `<host>-<pid>`.
+*/
+async function browserPid(context, profileDir) {
+	const browser = context.browser();
+	if (browser !== null) try {
+		const session = await browser.newBrowserCDPSession();
+		try {
+			const main = (await session.send("SystemInfo.getProcessInfo")).processInfo.find((item) => item.type === "browser");
+			if (main !== void 0 && main.id > 0) return main.id;
+		} finally {
+			await session.detach().catch(() => void 0);
+		}
+	} catch {}
+	if (process.platform === "win32") return void 0;
+	try {
+		const target = await readlink(join(profileDir, "SingletonLock"));
+		const pid = Number(target.slice(target.lastIndexOf("-") + 1));
+		return Number.isInteger(pid) && pid > 0 ? pid : void 0;
+	} catch {
+		return;
+	}
+}
+/**
+* Ask the browser to close; kill `pid` when that fails or takes longer than
+* `waitMs`, so a stuck browser does not keep the profile locked.
+*/
+async function closeOrKill(close, pid, waitMs = 1e3, kill = (target) => {
+	process.kill(target, "SIGKILL");
+}) {
+	let timer;
+	const outcome = await Promise.race([close().then(() => "closed", () => "failed"), new Promise((resolve) => {
+		timer = setTimeout(() => resolve("timeout"), waitMs);
+	})]);
+	clearTimeout(timer);
+	if (outcome === "closed") return "closed";
+	if (pid === void 0) return "stuck";
+	try {
+		kill(pid);
+	} catch {}
+	return "killed";
+}
 //#endregion
 //#region lib/types/service/control.js
 /**
@@ -4121,7 +4165,7 @@ const REPLIED = "继续";
 //#endregion
 //#region lib/types/service/extension.js
 const PACKAGE = "@mind-harness/mh-browser";
-const WORKER_MS = 5e3;
+const WORKER_MS = 15e3;
 /**
 * Directory of the unpacked extension shipped with this package.
 * Walks up from this module: sources live in `src/`, the bundle in `lib/`.
@@ -4156,6 +4200,10 @@ var Extension = class Extension {
 		const worker = await findWorker(context, loaded.id);
 		return new Extension(context, loaded.id, worker);
 	}
+	/** Address of a page shipped in the extension directory. */
+	pageUrl(name) {
+		return `chrome-extension://${this.id}/${name}`;
+	}
 	/** `chrome.tabs.create` in the named window. Returns the extension's tab id. */
 	async createTab(windowId, url) {
 		return (await this.worker()).evaluate(async (arg) => {
@@ -4231,13 +4279,21 @@ var Extension = class Extension {
 		return worker;
 	}
 };
-async function findWorker(context, id) {
-	const open = context.serviceWorkers().find((worker) => worker.url().includes(id));
+/** The extension's service worker. Looks once more after the wait times out. */
+async function findWorker(context, id, timeoutMs = WORKER_MS) {
+	const find = () => context.serviceWorkers().find((worker) => worker.url().includes(id));
+	const open = find();
 	if (open !== void 0) return open;
-	return context.waitForEvent("serviceworker", {
-		predicate: (worker) => worker.url().includes(id),
-		timeout: WORKER_MS
-	});
+	try {
+		return await context.waitForEvent("serviceworker", {
+			predicate: (worker) => worker.url().includes(id),
+			timeout: timeoutMs
+		});
+	} catch (error) {
+		const late = find();
+		if (late !== void 0) return late;
+		throw new Error(`the browser extension did not start within ${String(timeoutMs / 1e3)} s`, { cause: error });
+	}
 }
 //#endregion
 //#region lib/types/service/windows.js
@@ -4319,6 +4375,8 @@ var Windows = class {
 		this.browserSession = await browser.newBrowserCDPSession();
 		if (options.headed) this.extension = await Extension.load(context, this.browserSession);
 		const launched = context.pages().filter((page) => page.url() === "about:blank");
+		const extension = this.extension;
+		if (extension !== void 0) for (const page of launched) await page.goto(extension.pageUrl("welcome.html")).catch(() => void 0);
 		context.on("page", (page) => {
 			for (const blank of launched.splice(0)) blank.close().catch(() => void 0);
 			this.ingest(page);
@@ -5066,6 +5124,7 @@ let MhBrowserService = (() => {
 		windows = (__runInitializers(this, _instanceExtraInitializers), new Windows());
 		launchOptions;
 		context;
+		pids = /* @__PURE__ */ new WeakMap();
 		ready = Promise.resolve();
 		closing = false;
 		/** The browser is not running. The next call that needs a page starts it. */
@@ -5123,7 +5182,8 @@ let MhBrowserService = (() => {
 				this.agentViewOrigins.add(parsed.origin);
 			} catch {}
 		}
-		async open(owner, items, signal) {
+		async open(owner, items, signal, options) {
+			const reveal = options?.reveal === true;
 			return this.exclusive(owner, async () => {
 				await this.up(owner === PERSON_OWNER);
 				const views = [];
@@ -5135,7 +5195,7 @@ let MhBrowserService = (() => {
 						});
 						continue;
 					}
-					views.push(await this.openAddress(owner, item, signal));
+					views.push(await this.openAddress(owner, item, signal, reveal));
 				}
 				await this.enforceCap(owner);
 				return views;
@@ -5298,7 +5358,9 @@ let MhBrowserService = (() => {
 					await (this.context?.browser())?.close();
 				},
 				page: (owner, tab) => this.windows.get(owner, tab)?.page,
-				running: () => this.context !== void 0
+				running: () => this.context !== void 0,
+				pages: () => this.context?.pages() ?? [],
+				pid: () => this.context === void 0 ? void 0 : this.pids.get(this.context)
 			};
 		}
 		async act(owner, tab, op, signal) {
@@ -5503,7 +5565,7 @@ let MhBrowserService = (() => {
 		* A readable Agent View found on a tab this call opened is returned without
 		* the tab, and that tab is not listed as closed.
 		*/
-		async openAddress(owner, item, signal) {
+		async openAddress(owner, item, signal, reveal = false) {
 			if (!openableAddress(item.url)) return {
 				url: item.url,
 				reason: OPEN_ADDRESS_REASON
@@ -5517,7 +5579,7 @@ let MhBrowserService = (() => {
 				};
 				if (via !== void 0) return via;
 			}
-			const opened = await this.openOne(owner, item.url, item.waitFor, signal, item.snapshot === true);
+			const opened = await this.openOne(owner, item.url, item.waitFor, signal, item.snapshot === true, reveal);
 			this.keepAgentView(opened);
 			if (!wantsTab && opened.fresh && this.closable(owner, opened.live) && "representation" in opened.view && opened.view.representation.kind === "agent-view") {
 				await this.windows.closeTab(opened.live);
@@ -5686,12 +5748,13 @@ let MhBrowserService = (() => {
 				return body;
 			}
 		}
-		async openOne(owner, url, waitFor, signal, snapshot = false) {
+		async openOne(owner, url, waitFor, signal, snapshot = false, reveal = false) {
 			assertWaitFor(waitFor);
 			const reused = this.idleTabOn(owner, url);
 			const live = reused ?? await this.windows.openBlank(owner);
 			const fresh = reused === void 0;
 			const linked = this.control.link(owner, signal, live.id);
+			if (reveal && !linked.signal.aborted) await live.page.bringToFront().catch(() => void 0);
 			const nextDialog = this.windows.nextDialog(live);
 			try {
 				let seenDialog;
@@ -6365,6 +6428,8 @@ let MhBrowserService = (() => {
 		}
 		async start() {
 			const context = await launchBrowser(this.launchOptions);
+			const pid = await browserPid(context, this.launchOptions.profileDir);
+			if (pid !== void 0) this.pids.set(context, pid);
 			if (this.closing) {
 				await this.closeContext(context);
 				return;
@@ -6423,9 +6488,8 @@ let MhBrowserService = (() => {
 		}
 		/** Close a context the service is done with, and kill it if close stalls. */
 		async closeContext(context) {
-			const browser = context.browser();
-			await Promise.race([context.close().catch(() => void 0), new Promise((resolve) => setTimeout(resolve, 1e3))]);
-			(browser?.process?.())?.kill("SIGKILL");
+			await closeOrKill(() => context.close(), this.pids.get(context));
+			this.pids.delete(context);
 		}
 		refuseIfPerson(owner, tab) {
 			if (this.control.ofTab(owner, tab) !== "person") return;

package/lib/typert.host.js

@@ -44,7 +44,7 @@ const $remote = createTypertRemoteInfo({
         create: _mind_harness_mh_browser_mhBrowser_revealHandover_result$schema,
         jsonSchema: {"type":"object","properties":{"status":{"type":"string","enum":["revealed","not-handing-over","no-tab"]}},"required":["status"],"additionalProperties":true},
       },
-      sourceLocation: {"file":"packages/mh-browser/src/service/service.ts","line":619,"column":9},
+      sourceLocation: {"file":"packages/mh-browser/src/service/service.ts","line":624,"column":9},
     },
   },
 }, z)
@@ -75,7 +75,7 @@ export const TYPERT = {
           {
             "kind": "method",
             "name": "open",
-            "signature": "async open( owner: string, items: { url: string; waitFor?: WaitFor; snapshot?: boolean }[], signal?: AbortSignal, ): Promise<(PageView | PageError)[]>"
+            "signature": "async open( owner: string, items: { url: string; waitFor?: WaitFor; snapshot?: boolean }[], signal?: AbortSignal, options?: { reveal?: boolean }, ): Promise<(PageView | PageError)[]>"
           },
           {
             "kind": "method",
@@ -359,7 +359,7 @@ export const TYPERT = {
           },
           {
             "name": "TestHooks",
-            "declaration": "export interface TestHooks {\n    crash(): Promise<void>;\n    page(owner: string, tab: string): unknown;\n    running(): boolean;\n}"
+            "declaration": "export interface TestHooks {\n    crash(): Promise<void>;\n    page(owner: string, tab: string): unknown;\n    running(): boolean;\n    pages(): unknown[];\n    pid(): number | undefined;\n}"
           },
           {
             "name": "WaitFor",

package/lib/typert.remote-client.js

@@ -44,7 +44,7 @@ export const TYPERT_REMOTE = createTypertRemoteInfo({
         create: _mind_harness_mh_browser_mhBrowser_revealHandover_result$schema,
         jsonSchema: {"type":"object","properties":{"status":{"type":"string","enum":["revealed","not-handing-over","no-tab"]}},"required":["status"],"additionalProperties":true},
       },
-      sourceLocation: {"file":"packages/mh-browser/src/service/service.ts","line":619,"column":9},
+      sourceLocation: {"file":"packages/mh-browser/src/service/service.ts","line":624,"column":9},
     },
   },
 }, z)

package/lib/types/service/browser.d.ts

@@ -20,4 +20,15 @@ export declare function chromiumSandboxEnabled(browser: BrowserName, platform?:
  * A missing browser is reported and not downloaded.
  */
 export declare function launchBrowser(options: LaunchOptions): Promise<BrowserContext>;
+/**
+ * Process id of the browser's main process. Playwright gives none for a
+ * persistent context. Asks the browser first; on POSIX the profile's
+ * `SingletonLock` link names it as `<host>-<pid>`.
+ */
+export declare function browserPid(context: BrowserContext, profileDir: string): Promise<number | undefined>;
+/**
+ * Ask the browser to close; kill `pid` when that fails or takes longer than
+ * `waitMs`, so a stuck browser does not keep the profile locked.
+ */
+export declare function closeOrKill(close: () => Promise<void>, pid: number | undefined, waitMs?: number, kill?: (target: number) => void): Promise<'closed' | 'killed' | 'stuck'>;
 //# sourceMappingURL=browser.d.ts.map

package/lib/types/service/extension.d.ts

@@ -1,4 +1,4 @@
-import type { BrowserContext, CDPSession } from 'playwright';
+import type { BrowserContext, CDPSession, Worker } from 'playwright';
 /**
  * Directory of the unpacked extension shipped with this package.
  * Walks up from this module: sources live in `src/`, the bundle in `lib/`.
@@ -14,6 +14,8 @@ export declare class Extension {
     private workerPage;
     private constructor();
     static load(context: BrowserContext, browserSession: CDPSession): Promise<Extension>;
+    /** Address of a page shipped in the extension directory. */
+    pageUrl(name: string): string;
     /** `chrome.tabs.create` in the named window. Returns the extension's tab id. */
     createTab(windowId: number, url: string): Promise<number>;
     tabIdsIn(windowId: number): Promise<number[]>;
@@ -27,4 +29,6 @@ export declare class Extension {
     ungroup(tabId: number): Promise<void>;
     private worker;
 }
+/** The extension's service worker. Looks once more after the wait times out. */
+export declare function findWorker(context: BrowserContext, id: string, timeoutMs?: number): Promise<Worker>;
 //# sourceMappingURL=extension.d.ts.map

package/lib/types/service/service.d.ts

@@ -34,6 +34,7 @@ export declare class MhBrowserService extends TypertRemoteService {
     private readonly windows;
     private readonly launchOptions;
     private context;
+    private readonly pids;
     private ready;
     private closing;
     /** The browser is not running. The next call that needs a page starts it. */
@@ -58,7 +59,9 @@ export declare class MhBrowserService extends TypertRemoteService {
         url: string;
         waitFor?: WaitFor;
         snapshot?: boolean;
-    }[], signal?: AbortSignal): Promise<(PageView | PageError)[]>;
+    }[], signal?: AbortSignal, options?: {
+        reveal?: boolean;
+    }): Promise<(PageView | PageError)[]>;
     view(owner: string, tab: string, signal?: AbortSignal): Promise<PageView | PageError>;
     /** Pages whose current representation is already in the model's presence. */
     rememberPresence(owner: string, urls: string[]): void;

package/lib/types/service/test-hooks.d.ts

@@ -5,5 +5,9 @@ export interface TestHooks {
     page(owner: string, tab: string): unknown;
     /** A browser process is up. */
     running(): boolean;
+    /** Every page of the running browser, launch page included. */
+    pages(): unknown[];
+    /** The main browser process id recorded at start. */
+    pid(): number | undefined;
 }
 //# sourceMappingURL=test-hooks.d.ts.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mind-harness/mh-browser",
-  "version": "0.2.2",
+  "version": "0.2.3",
   "description": "Opens pages in a real browser for the agent, keeps chosen Agent Views in context, and hands the browser to you when a person is needed.",
   "license": "SEE LICENSE IN LICENSE",
   "files": [