@agentorg/computer 0.0.12 → 0.0.14

package/README.md

@@ -35,6 +35,8 @@ That prints the computer's `computerID`, its endpoints and its `accessToken`, on
 | `terminal` | Those, and `exec_command`, `write_stdin`, `list_sessions` |
 | `none`     | No `computer` endpoint; needs `--browser`                 |
 
+`read` answers a text file as numbered lines, and a PNG, JPEG, GIF or WebP of up to 3 MiB as the picture itself: an image an MCP client can pass on to a model. The browser's `browser_take_screenshot` answers an image too, of whatever size it comes to: a `fullPage` one of a long page can be many megabytes. A container made by an earlier version keeps the program it was made with, and refuses a picture as a binary file until it is killed and made again.
+
 `about` tells the agent what a person looking at the computer reads. It takes nothing and answers JSON text, `{"rows": [...]}`, each row a `label` and a `value`, with `"path": true` on a value that is a path and `"caution": true` on one to be read as a caution; a row that is neither has neither. The rows come in this order:
 
 | `label`     | `value`                                                                                   |
@@ -51,7 +53,7 @@ That prints the computer's `computerID`, its endpoints and its `accessToken`, on
 - **The last three** are a reading of its load taken over half a second when it is asked, in the same way the `load` surface reads. They come together, or not at all where a reading can't be taken.
 - **Not every computer has it.** One with no `computer` endpoint has no `about`. One in a container made by an earlier version keeps the program it was made with: `about` is listed among its tools and isn't there until it is killed and made again.
 
-The file tools stay inside the workspace. That's a check, not a boundary. Below `terminal`, the browser's `browser_run_code_unsafe` tool is hidden and refused, because it would otherwise be a terminal. The service enforces all of this, whatever the client does.
+The file tools stay inside the workspace. That's a check, not a boundary. Below `terminal`, the browser's `browser_run_code_unsafe` tool is hidden and refused, because it would otherwise be a terminal. The browser's endpoint takes one JSON-RPC request in each POST, as a JSON object, and refuses anything else with 400, a batch among them: it checks each request before passing it on, and passes on only what it read. The service enforces all of this, whatever the client does.
 
 `--browser` drives the Google Chrome installed on the machine, in a profile of the computer's own. `serve --browser-name` and `--browser-path` choose another. The browser has no debug port: a small host process starts it and shares it on a Unix socket in a folder only its user can enter, which the agent's tools and the live view both use. It lives as long as its computer does: a client that ends its session or goes away leaves the browser and its pages as they are, and the agent's `browser_close` is refused. Only an agent with `--access terminal` can stop it, with `browser_run_code_unsafe` or from its terminal.
 

package/dist/THIRD_PARTY_LICENSES.txt

@@ -1,4 +1,4 @@
-@agentorg/exec 0.0.6 (Apache-2.0)
+@agentorg/exec 0.0.7 (Apache-2.0)
 
 Apache License
                            Version 2.0, January 2004

package/dist/agentcomputer.js

@@ -3268,7 +3268,7 @@ var require_stream = __commonJS({
       };
       duplex._final = function(callback) {
         if (ws.readyState === ws.CONNECTING) {
-          ws.once("open", function open4() {
+          ws.once("open", function open5() {
             duplex._final(callback);
           });
           return;
@@ -3289,7 +3289,7 @@ var require_stream = __commonJS({
       };
       duplex._write = function(chunk, encoding, callback) {
         if (ws.readyState === ws.CONNECTING) {
-          ws.once("open", function open4() {
+          ws.once("open", function open5() {
             duplex._write(chunk, encoding, callback);
           });
           return;
@@ -27961,6 +27961,7 @@ var RUNS_CODE = "browser_run_code_unsafe";
 var MAX_BODY = 8 * 1024 * 1024;
 var START_MS = 3e4;
 var STOP_MS = 2e3;
+var ONE_REQUEST = "A body is one JSON-RPC request, as a JSON object: send each request in its own POST.";
 var FORWARDED = [
   "accept",
   "content-type",
@@ -28098,18 +28099,29 @@ var Browser = class {
       chunks.push(chunk);
     }
     const body = Buffer.concat(chunks);
-    let message = {};
+    let message;
     try {
-      if (body.length > 0) message = JSON.parse(body.toString("utf8"));
+      const read = body.length > 0 ? JSON.parse(body.toString("utf8")) : void 0;
+      if (typeof read === "object" && read !== null && !Array.isArray(read)) message = read;
     } catch {
     }
-    const tool = message.method === "tools/call" ? String(message.params?.name) : void 0;
+    if (body.length > 0 && !message) {
+      response.writeHead(400, { "Content-Type": "application/json" });
+      return void response.end(
+        JSON.stringify({
+          jsonrpc: "2.0",
+          id: null,
+          error: { code: -32600, message: ONE_REQUEST }
+        })
+      );
+    }
+    const tool = message?.method === "tools/call" ? String(message.params?.name) : void 0;
     if (tool && this.#hidden.has(tool)) {
       response.writeHead(200, { "Content-Type": "application/json" });
       return void response.end(
         JSON.stringify({
           jsonrpc: "2.0",
-          id: message.id ?? null,
+          id: message?.id ?? null,
           error: { code: -32602, message: `Unknown tool: ${tool}` }
         })
       );
@@ -28134,12 +28146,13 @@ var Browser = class {
       }).then((answer) => answer.text()).catch(() => {
       });
     if (tool === "browser_file_upload") this.gate?.agentAnsweredChooser();
+    if (message) headers["content-type"] = "application/json";
     const upstream = await fetch(`${origin}/mcp`, {
       method: request.method,
       headers,
-      ...body.length > 0 ? { body } : {}
+      ...message ? { body: JSON.stringify(message) } : {}
     });
-    if (message.method === "tools/list" && this.#hidden.size > 0) {
+    if (message?.method === "tools/list" && this.#hidden.size > 0) {
       const text3 = withoutHidden(await upstream.text(), this.#hidden);
       return sendAnswer(new Response(text3, upstream), response);
     }
@@ -28169,12 +28182,13 @@ import { STATUS_CODES, ServerResponse, createServer } from "node:http";
 import { Http2ServerRequest, constants } from "node:http2";
 import { Readable as Readable2 } from "node:stream";
 import { randomBytes as randomBytes3 } from "node:crypto";
-import { lstat as lstat2, open as open2, readdir as readdir2, rename as rename2, rm as rm2, stat as stat2 } from "node:fs/promises";
+import { lstat as lstat2, open as open3, readdir as readdir2, rename as rename2, rm as rm2, stat as stat2 } from "node:fs/promises";
 import { dirname as dirname2, join as join3 } from "node:path";
 import { randomBytes as randomBytes2 } from "node:crypto";
 import {
   chmod,
   lstat,
+  open as open2,
   readdir,
   readFile,
   realpath,
@@ -34318,7 +34332,7 @@ function bucketFor2(state, inst) {
   return bucket;
 }
 var handoff2;
-var open3 = [];
+var open4 = [];
 var memo3 = {
   alloc(_inst, payload, empty) {
     const bucket = handoff2;
@@ -34327,7 +34341,7 @@ var memo3 = {
     handoff2 = void 0;
     const entry = { value: empty, issues: null };
     bucket.set(payload.value, entry);
-    open3.push(entry);
+    open4.push(entry);
     return empty;
   },
   guard(inst) {
@@ -34398,10 +34412,10 @@ var memo3 = {
           return payload;
         }
         handoff2 = bucket;
-        const depth = open3.length;
+        const depth = open4.length;
         const result = base(payload, ctx);
         handoff2 = void 0;
-        const entry = open3.length > depth ? open3.pop() : void 0;
+        const entry = open4.length > depth ? open4.pop() : void 0;
         if (result instanceof Promise) {
           return result.then((r) => {
             if (entry)
@@ -66158,6 +66172,21 @@ var MAX_READ_CHARS = 28e3;
 var MAX_LINE_CHARS = 2e3;
 var MAX_ENTRIES = 1e3;
 var MAX_EDIT_BYTES = 8 * 1024 * 1024;
+var MAX_PICTURE_BYTES = 3 * 1024 * 1024;
+async function pictureKind(path) {
+  const file2 = await open2(path);
+  try {
+    const { buffer, bytesRead } = await file2.read(Buffer.alloc(12), 0, 12, 0);
+    const start = buffer.toString("latin1", 0, bytesRead);
+    if (start.startsWith("\x89PNG\r\n\n")) return "image/png";
+    if (start.startsWith("\xFF\xD8\xFF")) return "image/jpeg";
+    if (start.startsWith("GIF87a") || start.startsWith("GIF89a")) return "image/gif";
+    if (start.startsWith("RIFF") && start.slice(8) === "WEBP") return "image/webp";
+    return void 0;
+  } finally {
+    await file2.close();
+  }
+}
 var reason = (error62) => {
   const { code, path } = error62;
   if (code === "ENOENT") return `No such file or folder: ${path}`;
@@ -66213,9 +66242,19 @@ function createFiles(root) {
     return bytes.toString("utf8");
   };
   return {
-    /** Numbered lines of a file. */
+    /** Numbered lines of a file, or the file itself where it is a picture. */
     read: (input2) => refusing(async () => {
-      const lines = (await text22(await at(input2.path), input2.path)).split("\n");
+      const path = await at(input2.path);
+      const { size } = await aFile(path, input2.path);
+      const mimeType = await pictureKind(path);
+      if (mimeType) {
+        if (size > MAX_PICTURE_BYTES)
+          throw new ExecRefused(
+            `${input2.path} is too large to read as a picture: ${size.toLocaleString("en-US")} bytes, and the most is ${MAX_PICTURE_BYTES.toLocaleString("en-US")}.`
+          );
+        return { mimeType, bytes: await readFile(path) };
+      }
+      const lines = (await text22(path, input2.path)).split("\n");
       const first2 = Math.max(input2.offset ?? 1, 1);
       const shown = lines.slice(first2 - 1, first2 - 1 + (input2.limit ?? DEFAULT_LINES)).map((line, i) => `${first2 + i}	${cut(line, MAX_LINE_CHARS)}`);
       const last = first2 + shown.length - 1;
@@ -66425,7 +66464,7 @@ function createHostFiles(root, level) {
         throw new ExecRefused("A read starts at 0 or later and takes 1 to 1,048,576 bytes.");
       const real = await inside(root, path);
       const found = await aFile(real, path);
-      const file2 = await open2(real, "r");
+      const file2 = await open3(real, "r");
       try {
         const bytes = Buffer.alloc(Math.max(0, Math.min(length, found.size - offset)));
         const { bytesRead } = await file2.read(bytes, 0, bytes.length, offset);
@@ -66454,7 +66493,7 @@ function createHostFiles(root, level) {
       if (before?.isDirectory()) throw new ExecRefused(`${path} is a folder.`);
       await stat2(dirname2(target));
       const temporary = join3(dirname2(target), `.agentexec-${randomBytes3(6).toString("hex")}`);
-      const file2 = await open2(temporary, "wx");
+      const file2 = await open3(temporary, "wx");
       const discard = async () => {
         await file2.close().catch(() => {
         });
@@ -66527,6 +66566,12 @@ var answering = async (work) => {
     return refused(error62);
   }
 };
+var pictured = (path, { mimeType, bytes }) => ({
+  content: [
+    { type: "text", text: `${path}: ${mimeType}` },
+    { type: "image", mimeType, data: bytes.toString("base64") }
+  ]
+});
 var saying = async (work) => {
   try {
     return text(await work());
@@ -66626,7 +66671,7 @@ function createServer2(root, sessions, version22, level, extraTools = []) {
   server.registerTool(
     "read",
     {
-      description: "Reads a text file as numbered lines.",
+      description: "Reads a text file as numbered lines, or shows a PNG, JPEG, GIF or WebP of up to 3 MiB as an image; offset and limit are for text.",
       inputSchema: {
         path: external_exports.string().min(1).describe("Inside the workspace; relative to it, or absolute."),
         offset: external_exports.number().int().positive().optional().describe("The first line, from 1."),
@@ -66634,7 +66679,14 @@ function createServer2(root, sessions, version22, level, extraTools = []) {
       },
       annotations: READ_ONLY
     },
-    (input2) => saying(() => files.read(input2))
+    async (input2) => {
+      try {
+        const read = await files.read(input2);
+        return typeof read === "string" ? text(read) : pictured(input2.path, read);
+      } catch (error62) {
+        return refused(error62);
+      }
+    }
   );
   server.registerTool(
     "list",
@@ -66677,7 +66729,7 @@ function createServer2(root, sessions, version22, level, extraTools = []) {
     );
   return server;
 }
-var VERSION = "0.0.6";
+var VERSION = "0.0.7";
 function createAgentExec(options) {
   const root = realpathSync(resolve3(options.workspace));
   const level = options.level ?? "read";
@@ -68093,7 +68145,7 @@ async function forward(url3, authorization, request, response) {
 }
 
 // src/handlers.ts
-var VERSION2 = "0.0.12";
+var VERSION2 = "0.0.14";
 function createHandlers(computers) {
   return {
     getInfo: async () => ({ version: VERSION2, isolation: computers.isolation }),
@@ -68677,8 +68729,8 @@ async function openLive(session, onMessage) {
 }
 var Terminals = class {
   #open;
-  constructor(open4) {
-    this.#open = open4;
+  constructor(open5) {
+    this.#open = open5;
   }
   /**
    * Each terminal as it starts, writes and exits: those already there first, each with the end of
@@ -85058,7 +85110,7 @@ var OPENAPI = {
   "openapi": "3.1.0",
   "info": {
     "title": "AgentComputer",
-    "version": "0.0.12",
+    "version": "0.0.14",
     "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",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agentorg/computer",
-  "version": "0.0.12",
+  "version": "0.0.14",
   "description": "Computers for agents, as a service: a terminal, files and a browser over MCP",
   "license": "Apache-2.0",
   "repository": {
@@ -31,7 +31,7 @@
     "playwright-core": "1.65.0-alpha-2026-10-30"
   },
   "devDependencies": {
-    "@agentorg/exec": "0.0.6",
+    "@agentorg/exec": "0.0.7",
     "@biomejs/biome": "2.6.5",
     "@modelcontextprotocol/client": "2.4.0",
     "@modelcontextprotocol/node": "2.1.2",