@agentorg/computer 0.0.5 → 0.0.7

package/README.md

@@ -139,7 +139,7 @@ A computer at `--access terminal` has two surfaces you can open live, as a WebSo
 
 A computer with `--browser` has a third:
 
-- **`browser`** (protocol `screencast@1`): its browser, on the same page the agent works on. Binary messages are JPEG frames of the page shown. Text messages are JSON: the service sends `tabs`, `viewport`, `dialog`, `file-chooser`, `choice` and `driver`, and an `operate` socket sends `mouse`, `key`, `text`, `dialog`, `files`, `choice`, `tab`, `navigate` and `release`. `openapi.yaml` has each one's fields. A page's own dialogs, file choosers and select lists aren't in the frames, so they arrive as events for you to show and answer. Operating it makes you the driver: the agent's browser calls wait up to 25 seconds while you are, then answer that a person has or had the browser. You stop being the driver when you send `release`, close the socket, or send nothing for 60 seconds.
+- **`browser`** (protocol `screencast@1`): its browser, on the same page the agent works on. Binary messages are JPEG frames of the page shown. Text messages are JSON: the service sends `tabs`, `viewport`, `dialog`, `file-chooser`, `choice` and `driver`, and an `operate` socket sends `mouse`, `key`, `text`, `dialog`, `files`, `choice`, `tab`, `navigate` and `release`. `openapi.yaml` has each one's fields. A page's own dialogs, file choosers and select lists aren't in the frames, so they arrive as events for you to show and answer. Opening it to operate makes you the driver: the agent's browser calls wait up to 25 seconds while you are, then answer that a person has or had the browser. You stop being the driver when you send `release`, close the socket, or send nothing for 60 seconds, and sending input takes it again.
 
 `POST /v1/computers/{computerID}/live` with `{ "surface": "terminal", "mode": "operate" }` gives a ticket: a `url` to open once within a minute, which needs no token. A request with an `Origin` header is refused unless `serve --allow-origin` names that origin, so a page can't open one. The SDK does the same with `computer.pty.connect({ onData })` and `computer.commands.watch(onEvent)`, and the CLI with `agentcomputer live <id> --surface terminal --mode operate`.
 

package/dist/agentcomputer.js

@@ -3759,7 +3759,6 @@ import { isMainThread as isMainThread2 } from "node:worker_threads";
 
 // src/browser-host.ts
 import { mkdtempSync, rmSync } from "node:fs";
-import { tmpdir } from "node:os";
 import { join, sep } from "node:path";
 
 // src/playwright.ts
@@ -3773,8 +3772,9 @@ var playwright = () => {
 // src/browser-host.ts
 var sandboxFor = (options) => options.sandbox ?? (process.platform !== "linux" || !options.executable && !["chromium", "chrome-for-testing"].includes(options.channel ?? "chrome"));
 async function hostBrowser(options) {
-  const run3 = mkdtempSync(join(tmpdir(), "agentcomputer-"));
+  const run3 = mkdtempSync(join("/tmp", "agentcomputer-"));
   process.env.XDG_RUNTIME_DIR = run3;
+  process.env.TMPDIR = run3;
   const context = await playwright().chromium.launchPersistentContext(options.profile, {
     headless: true,
     chromiumSandbox: sandboxFor(options),
@@ -66315,6 +66315,7 @@ async function startScreencast(page, onFrame) {
 // src/browser-live.ts
 var WAIT_MS = 25e3;
 var IDLE_MS = 6e4;
+var TITLE_CHECK_MS = 1e3;
 var STEP_MS = 2e3;
 var MAX_BUFFERED = 4 * 1024 * 1024;
 var sleep3 = (ms) => new Promise((done) => setTimeout(done, ms));
@@ -66337,6 +66338,10 @@ var BrowserLive = class {
   #lastInput = 0;
   #agentWaiting = 0;
   #answeredChooser = false;
+  /** The tabs last sent, to send them again only when something about them changed. */
+  #lastTabs = "";
+  #titles;
+  #checking = false;
   #steps = Promise.resolve();
   constructor(endpoint) {
     this.#endpoint = endpoint;
@@ -66369,6 +66374,10 @@ var BrowserLive = class {
   // ---- The person's side
   open(ws, mode) {
     this.#sockets.set(ws, mode);
+    if (mode === "operate") {
+      this.#driver = ws;
+      this.#lastInput = Date.now();
+    }
     ws.on("close", () => this.#left(ws));
     ws.on("message", (data, binary) => {
       if (binary || this.#sockets.get(ws) !== "operate") return;
@@ -66418,6 +66427,7 @@ var BrowserLive = class {
         void this.#show(page);
       });
       await this.#show(context.pages().at(-1) ?? await context.newPage());
+      this.#titles = setInterval(() => void this.#checkTabs(), TITLE_CHECK_MS);
       return viewer;
     })();
     await this.#viewer;
@@ -66425,6 +66435,9 @@ var BrowserLive = class {
   async #end() {
     const viewer = this.#viewer;
     this.#viewer = void 0;
+    clearInterval(this.#titles);
+    this.#titles = void 0;
+    this.#lastTabs = "";
     await this.#stopCast?.();
     this.#stopCast = void 0;
     this.#shown = void 0;
@@ -66499,7 +66512,7 @@ var BrowserLive = class {
     }
     await this.#tabs();
   }
-  async #tabs(only) {
+  async #tabList() {
     const tabs = await Promise.all(
       [...this.#ids].map(async ([page, id]) => ({
         id,
@@ -66507,14 +66520,28 @@ var BrowserLive = class {
         title: await within(page.title(), STEP_MS).catch(() => void 0) ?? ""
       }))
     );
-    const message = {
-      type: "tabs",
-      tabs,
-      shown: (this.#shown && this.#ids.get(this.#shown)) ?? null
-    };
+    return { type: "tabs", tabs, shown: (this.#shown && this.#ids.get(this.#shown)) ?? null };
+  }
+  async #tabs(only) {
+    const message = await this.#tabList();
+    this.#lastTabs = JSON.stringify(message);
     if (only) this.#send(only, message);
     else this.#broadcast(message);
   }
+  /** Send the tabs again if anything about them changed, such as a title. */
+  async #checkTabs() {
+    if (this.#checking) return;
+    this.#checking = true;
+    try {
+      const message = await this.#tabList();
+      const text2 = JSON.stringify(message);
+      if (text2 === this.#lastTabs) return;
+      this.#lastTabs = text2;
+      this.#broadcast(message);
+    } finally {
+      this.#checking = false;
+    }
+  }
   /** Each step in order, none waiting on another for long: a dialog blocks the page's input. */
   #step(work) {
     this.#steps = this.#steps.then(() => within(work(), STEP_MS)).catch(() => {
@@ -67185,7 +67212,7 @@ async function forward(url3, authorization, request, response) {
 }
 
 // src/handlers.ts
-var VERSION2 = "0.0.5";
+var VERSION2 = "0.0.7";
 function createHandlers(computers) {
   return {
     getInfo: async () => ({ version: VERSION2, isolation: computers.isolation }),
@@ -84160,7 +84187,7 @@ var OPENAPI = {
   "openapi": "3.1.0",
   "info": {
     "title": "AgentComputer",
-    "version": "0.0.5",
+    "version": "0.0.7",
     "description": "Computers for agents. This file is the source of truth: the HTTP API, the MCP control tools and the SDK are all generated from it. An operation with `x-mcp-tool` is also an MCP tool of that name.",
     "license": {
       "name": "Apache-2.0",
@@ -84824,7 +84851,7 @@ var OPENAPI = {
           "terminal",
           "browser"
         ],
-        "description": "commands: the agent's commands as they run, read-only, in commands@1. terminal: a shell of your own in its workspace, as its user, in pty@1. browser: its browser, in screencast@1: binary messages are JPEG frames of the page shown, text messages are ScreencastEvent one way and ScreencastInput the other. Operating it makes you the driver: the agent's browser calls wait up to 25 seconds while you are, then answer that a person has or had it. You stop being the driver when you send release, close the socket, or send nothing for 60 seconds."
+        "description": "commands: the agent's commands as they run, read-only, in commands@1. terminal: a shell of your own in its workspace, as its user, in pty@1. browser: its browser, in screencast@1: binary messages are JPEG frames of the page shown, text messages are ScreencastEvent one way and ScreencastInput the other. Opening it to operate makes you the driver: the agent's browser calls wait up to 25 seconds while you are, then answer that a person has or had it. You stop being the driver when you send release, close the socket, or send nothing for 60 seconds, and sending input takes it again."
       },
       "LiveMode": {
         "type": "string",
@@ -84915,7 +84942,7 @@ var OPENAPI = {
         "oneOf": [
           {
             "type": "object",
-            "description": "The open tabs, and which one the frames are of.",
+            "description": "The open tabs, and which one the frames are of, sent again whenever a tab opens, closes, navigates or changes its title.",
             "required": [
               "type",
               "tabs",
@@ -85846,7 +85873,7 @@ var OPERATIONS = [
                       "terminal",
                       "browser"
                     ],
-                    "description": "commands: the agent's commands as they run, read-only, in commands@1. terminal: a shell of your own in its workspace, as its user, in pty@1. browser: its browser, in screencast@1: binary messages are JPEG frames of the page shown, text messages are ScreencastEvent one way and ScreencastInput the other. Operating it makes you the driver: the agent's browser calls wait up to 25 seconds while you are, then answer that a person has or had it. You stop being the driver when you send release, close the socket, or send nothing for 60 seconds."
+                    "description": "commands: the agent's commands as they run, read-only, in commands@1. terminal: a shell of your own in its workspace, as its user, in pty@1. browser: its browser, in screencast@1: binary messages are JPEG frames of the page shown, text messages are ScreencastEvent one way and ScreencastInput the other. Opening it to operate makes you the driver: the agent's browser calls wait up to 25 seconds while you are, then answer that a person has or had it. You stop being the driver when you send release, close the socket, or send nothing for 60 seconds, and sending input takes it again."
                   },
                   "protocol": {
                     "type": "string"
@@ -86386,7 +86413,7 @@ var OPERATIONS = [
             "terminal",
             "browser"
           ],
-          "description": "commands: the agent's commands as they run, read-only, in commands@1. terminal: a shell of your own in its workspace, as its user, in pty@1. browser: its browser, in screencast@1: binary messages are JPEG frames of the page shown, text messages are ScreencastEvent one way and ScreencastInput the other. Operating it makes you the driver: the agent's browser calls wait up to 25 seconds while you are, then answer that a person has or had it. You stop being the driver when you send release, close the socket, or send nothing for 60 seconds."
+          "description": "commands: the agent's commands as they run, read-only, in commands@1. terminal: a shell of your own in its workspace, as its user, in pty@1. browser: its browser, in screencast@1: binary messages are JPEG frames of the page shown, text messages are ScreencastEvent one way and ScreencastInput the other. Opening it to operate makes you the driver: the agent's browser calls wait up to 25 seconds while you are, then answer that a person has or had it. You stop being the driver when you send release, close the socket, or send nothing for 60 seconds, and sending input takes it again."
         },
         "mode": {
           "allOf": [
@@ -86432,7 +86459,7 @@ var OPERATIONS = [
             "terminal",
             "browser"
           ],
-          "description": "commands: the agent's commands as they run, read-only, in commands@1. terminal: a shell of your own in its workspace, as its user, in pty@1. browser: its browser, in screencast@1: binary messages are JPEG frames of the page shown, text messages are ScreencastEvent one way and ScreencastInput the other. Operating it makes you the driver: the agent's browser calls wait up to 25 seconds while you are, then answer that a person has or had it. You stop being the driver when you send release, close the socket, or send nothing for 60 seconds."
+          "description": "commands: the agent's commands as they run, read-only, in commands@1. terminal: a shell of your own in its workspace, as its user, in pty@1. browser: its browser, in screencast@1: binary messages are JPEG frames of the page shown, text messages are ScreencastEvent one way and ScreencastInput the other. Opening it to operate makes you the driver: the agent's browser calls wait up to 25 seconds while you are, then answer that a person has or had it. You stop being the driver when you send release, close the socket, or send nothing for 60 seconds, and sending input takes it again."
         },
         "mode": {
           "type": "string",

package/dist/types/generated/schema.d.ts

@@ -248,7 +248,7 @@ export interface components {
             };
         };
         /**
-         * @description commands: the agent's commands as they run, read-only, in commands@1. terminal: a shell of your own in its workspace, as its user, in pty@1. browser: its browser, in screencast@1: binary messages are JPEG frames of the page shown, text messages are ScreencastEvent one way and ScreencastInput the other. Operating it makes you the driver: the agent's browser calls wait up to 25 seconds while you are, then answer that a person has or had it. You stop being the driver when you send release, close the socket, or send nothing for 60 seconds.
+         * @description commands: the agent's commands as they run, read-only, in commands@1. terminal: a shell of your own in its workspace, as its user, in pty@1. browser: its browser, in screencast@1: binary messages are JPEG frames of the page shown, text messages are ScreencastEvent one way and ScreencastInput the other. Opening it to operate makes you the driver: the agent's browser calls wait up to 25 seconds while you are, then answer that a person has or had it. You stop being the driver when you send release, close the socket, or send nothing for 60 seconds, and sending input takes it again.
          * @enum {string}
          */
         Surface: "commands" | "terminal" | "browser";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agentorg/computer",
-  "version": "0.0.5",
+  "version": "0.0.7",
   "description": "Computers for agents, as a service: a terminal, files and a browser over MCP",
   "license": "Apache-2.0",
   "repository": {