ask-pro 0.2.87 → 0.2.93

package/SKILL.md

@@ -11,7 +11,9 @@ The calling agent still owns the work. Use Pro for judgment, architecture, risk
 
 ## Running ask-pro
 
-On a headless Linux host, set `CHROME_PATH` if Chrome is not discovered, and
+On a display-less Linux host, install Xvfb (`sudo apt install xvfb`); ask-pro
+manages a private display for headed Chrome during login and queries. Set
+`CHROME_PATH` if Chrome is not discovered, and
 `ASK_PRO_LOGIN_HOST` to the host's private Tailscale IP. Before the first query,
 run `ask-pro --login` and give the human its `login_url`. Keep that command alive
 while they sign in from their desktop or phone. The link expires after 15 minutes

package/dist/src/ask-pro/loginPage.js

@@ -1,4 +1,5 @@
-// The browser relays human input only. Never persist input, frames, or the access key.
+// The browser relays human input only. Never persist input or frames.
+// Keep the short-lived access key in the URL fragment so reload preserves access.
 export const loginPage = `<!doctype html>
 <html lang="en"><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1">
 <title>ask-pro sign in</title>
@@ -8,7 +9,7 @@ export const loginPage = `<!doctype html>
 <p id="status" role="status">Connecting…</p><img id="screen" tabindex="0" alt="Remote browser. Click a field, then type, or use the text box below.">
 <form><input id="text" type="password" autocomplete="off" aria-label="Text for selected browser field" placeholder="Type into selected field"><button>Send text</button><button type="button" id="enter">Enter</button><button type="button" id="tab">Tab</button><button type="button" id="backspace">Backspace</button><button type="button" id="up">Scroll up</button><button type="button" id="down">Scroll down</button></form></main>
 <script>
-const token=location.hash.slice(1);history.replaceState(null,'',location.pathname);
+const token=location.hash.slice(1);
 const screen=document.querySelector('#screen'),tabs=document.querySelector('#tabs'),status=document.querySelector('#status'),text=document.querySelector('#text');
 let stopped=false,frameUrl,inputQueue=Promise.resolve();
 async function api(route,body){const r=await fetch(route,{method:body===undefined?'GET':'POST',headers:{Authorization:'Bearer '+token,'Content-Type':'application/json'},body:body===undefined?undefined:JSON.stringify(body),cache:'no-store'});if(!r.ok)throw Error(r.status===401?'Link expired. Start ask-pro --login again.':'Connection failed. Retry or refresh tabs.');return r;}

package/dist/src/ask-pro/remoteLogin.js

@@ -80,7 +80,29 @@ export async function startLoginServer(options) {
             }
             if (req.method === "GET" && route.pathname === "/frame") {
                 const c = await connect(route.searchParams.get("target") ?? "");
-                const frame = await c.Page.captureScreenshot({ format: "jpeg", quality: 70 });
+                // Reload cancels the image request. Release its pending CDP capture so
+                // it cannot hold the shared request queue and freeze the new viewer.
+                const cancel = () => {
+                    if (!res.writableEnded) {
+                        if (client === c) {
+                            client = undefined;
+                            targetId = undefined;
+                        }
+                        void c.close().catch(() => undefined);
+                    }
+                };
+                if (res.destroyed) {
+                    cancel();
+                    return;
+                }
+                res.once("close", cancel);
+                let frame;
+                try {
+                    frame = await c.Page.captureScreenshot({ format: "jpeg", quality: 70 });
+                }
+                finally {
+                    res.removeListener("close", cancel);
+                }
                 res.setHeader("Content-Type", "image/jpeg");
                 res.end(Buffer.from(frame.data, "base64"));
                 return;
@@ -222,7 +244,9 @@ export async function runRemoteLogin(onReady, savedProfile) {
         if (peers !== 0)
             throw Error("Browser busy. Wait for ask-pro runs to finish before --login.");
         const chrome = (await maybeReuseRunningChrome(profile, () => { })) ??
-            (await launchChrome(resolveBrowserConfig({ manualLogin: true, headless: true }), profile, () => { }, { ignoreLockId: lock.lockId }));
+            (await launchChrome(resolveBrowserConfig({ manualLogin: true }), profile, () => { }, {
+                ignoreLockId: lock.lockId,
+            }));
         releaseChromeProcessHandle(chrome);
         if (chrome.pid)
             await writeChromePid(profile, chrome.pid);

package/dist/src/browser/chromeLifecycle.js

@@ -12,6 +12,8 @@ import { clearChromeLaunchStarting, clearIdleBrowserProfileCaches, cleanupStaleP
 import { delay } from "./utils.js";
 import { formatElapsed } from "./format.js";
 import { withTimeout } from "./reattachHelpers.js";
+import { isHeadlessLinux } from "./config.js";
+import { spawnChromeWithVirtualDisplay } from "./linuxChrome.js";
 const execFileAsync = promisify(execFile);
 const CPU_THROTTLING_OVERRIDE_FLAGS = new Set([
     "--disable-backgrounding-occluded-windows",
@@ -747,7 +749,9 @@ export function spawnWindowsProcessOutsideJob(command, args, options) {
 }
 const spawnManagedChrome = ((command, args, options) => process.platform === "win32"
     ? spawnWindowsProcessOutsideJob(command, args, options)
-    : spawn(command, args, options));
+    : isHeadlessLinux() && !args.some((arg) => arg.startsWith("--headless"))
+        ? spawnChromeWithVirtualDisplay(command, args, options)
+        : spawn(command, args, options));
 async function launchManagedChrome({ chromeFlags, chromePath, userDataDir, host, requestedPort, }) {
     const launcher = new Launcher({
         chromePath: chromePath ?? undefined,

package/dist/src/browser/config.js

@@ -79,7 +79,7 @@ export function resolveBrowserConfig(config) {
         cookieSyncWaitMs: config?.cookieSyncWaitMs ?? DEFAULT_BROWSER_CONFIG.cookieSyncWaitMs,
         inlineCookies: config?.inlineCookies ?? DEFAULT_BROWSER_CONFIG.inlineCookies,
         inlineCookiesSource: config?.inlineCookiesSource ?? DEFAULT_BROWSER_CONFIG.inlineCookiesSource,
-        headless: config?.headless ?? isHeadlessLinux(),
+        headless: config?.headless ?? false,
         keepBrowser: config?.keepBrowser ?? DEFAULT_BROWSER_CONFIG.keepBrowser,
         hideWindow: config?.hideWindow ?? DEFAULT_BROWSER_CONFIG.hideWindow,
         startMinimized: config?.startMinimized ?? DEFAULT_BROWSER_CONFIG.startMinimized,

package/dist/src/browser/index.js

@@ -3,7 +3,7 @@ import path from "node:path";
 import os from "node:os";
 import net from "node:net";
 import CDP from "chrome-remote-interface";
-import { DEFAULT_BROWSER_CONFIG, resolveBrowserConfig } from "./config.js";
+import { DEFAULT_BROWSER_CONFIG, resolveBrowserConfig, isHeadlessLinux } from "./config.js";
 import { launchChrome, registerTerminationHooks, hideChromeWindow, connectToRemoteChrome, connectWithNewTab, closeTab, closeRemoteChromeTarget, closeChromeGracefully, listRemoteChromeTargets, maybeReuseRunningChrome, releaseChromeProcessHandle, restoreChromeWindowByPid, shouldLaunchChromeMinimized, } from "./chromeLifecycle.js";
 import { syncCookies } from "./cookies.js";
 import { navigateToChatGPT, navigateToPromptReadyWithFallback, ensureNotBlocked, ensureLoggedIn, ensurePromptReady, installJavaScriptDialogAutoDismissal, ensureModelSelection, clearPromptComposer, waitForAssistantResponse, captureAssistantMarkdown, clearComposerAttachments, uploadAttachmentFile, waitForAttachmentCompletion, waitForUserTurnAttachments, readAssistantSnapshot, } from "./pageActions.js";
@@ -1346,7 +1346,7 @@ export async function runBrowserMode(options) {
                 ? "ChatGPT login required; leaving browser open so you can sign in."
                 : "Cloudflare challenge detected; leaving browser open so you can complete the check.");
             logger(`Reuse this browser profile with: ${reuseProfileHint}`);
-            if (isLoginRequired && manualLogin) {
+            if (isLoginRequired && manualLogin && !isHeadlessLinux()) {
                 const recoveryRuntime = client?.Runtime;
                 if (!recoveryRuntime) {
                     throw normalizedError;

package/dist/src/browser/linuxChrome.js

@@ -0,0 +1,82 @@
+import { spawn, spawnSync } from "node:child_process";
+import { fileURLToPath } from "node:url";
+import { randomBytes } from "node:crypto";
+import { mkdtemp, writeFile, rm } from "node:fs/promises";
+import os from "node:os";
+import path from "node:path";
+// A separate supervisor outlives the controller when a browser is kept for resume.
+// Its PID represents the managed browser lifetime, including its private display.
+export function spawnChromeWithVirtualDisplay(command, args, options) {
+    const check = spawnSync("Xvfb", ["-help"], { stdio: "ignore" });
+    if (check.error)
+        throw Error("Xvfb is required on Linux without a display. Run: sudo apt install xvfb");
+    return spawn(process.execPath, [fileURLToPath(import.meta.url), "--supervise", command, ...args], options);
+}
+async function supervise() {
+    const directory = await mkdtemp(path.join(os.tmpdir(), "ask-pro-display-"));
+    const authority = path.join(directory, "Xauthority");
+    const cookie = randomBytes(16);
+    const writeAuthority = async (number) => {
+        const fields = [
+            Buffer.alloc(0),
+            Buffer.from(number),
+            Buffer.from("MIT-MAGIC-COOKIE-1"),
+            cookie,
+        ];
+        const records = fields.map((field) => {
+            const length = Buffer.alloc(2);
+            length.writeUInt16BE(field.length);
+            return Buffer.concat([length, field]);
+        });
+        await writeFile(authority, Buffer.concat([Buffer.from([255, 255]), ...records]), {
+            mode: 0o600,
+        });
+    };
+    await writeAuthority("");
+    const display = spawn("Xvfb", ["-displayfd", "3", "-screen", "0", "1280x800x24", "-nolisten", "tcp", "-auth", authority], {
+        stdio: ["ignore", "ignore", "inherit", "pipe"],
+    });
+    let chrome;
+    const stop = () => {
+        if (chrome)
+            chrome.kill("SIGTERM");
+        else
+            display.kill();
+    };
+    process.on("SIGTERM", stop);
+    process.on("SIGINT", stop);
+    let timer;
+    try {
+        const number = await new Promise((resolve, reject) => {
+            timer = setTimeout(() => reject(Error("Xvfb did not start within 10 seconds")), 10_000);
+            let output = "";
+            display.once("error", reject);
+            display.once("exit", () => reject(Error("Xvfb exited before Chrome started")));
+            display.stdio[3].on("data", (chunk) => {
+                output += String(chunk);
+                if (/^\d+\n$/.test(output))
+                    resolve(output.trim());
+            });
+        }).finally(() => clearTimeout(timer));
+        await writeAuthority(number);
+        chrome = spawn(process.argv[3], process.argv.slice(4), {
+            env: { ...process.env, DISPLAY: `:${number}`, XAUTHORITY: authority },
+            stdio: "inherit",
+        });
+        display.once("exit", () => chrome?.kill("SIGTERM"));
+        process.exitCode = await new Promise((resolve, reject) => {
+            chrome.once("error", reject);
+            chrome.once("exit", (code) => resolve(code ?? 1));
+        });
+    }
+    finally {
+        display.kill();
+        await rm(directory, { recursive: true, force: true });
+    }
+}
+if (process.argv[2] === "--supervise") {
+    await supervise().catch((error) => {
+        console.error(error.message);
+        process.exitCode = 1;
+    });
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ask-pro",
-  "version": "0.2.87",
+  "version": "0.2.93",
   "description": "Browser-backed ChatGPT Pro escalation for hard engineering questions.",
   "keywords": [
     "agents",
@@ -29,6 +29,7 @@
     "dist/src/browser/attachRunning.js",
     "dist/src/browser/chatgptModelCatalog.js",
     "dist/src/browser/chromeLifecycle.js",
+    "dist/src/browser/linuxChrome.js",
     "dist/src/browser/config.js",
     "dist/src/browser/constants.js",
     "dist/src/browser/cookies.js",