@inetafrica/open-claudia 3.5.26 → 3.5.28

package/CHANGELOG.md

@@ -1,5 +1,12 @@
 # Changelog
 
+## 3.5.28
+
+The Browser app is always on: opening it while the bot is not browsing opens the bot's browser, instead of a dead end (Sumeet, 10 Oct: "half the time i have to ask you switch on the browser"; 11 Oct 07:43 EAT network-mon's Browser showed only "isn't open right now"). A canary build for the Browser canary bot, on 3.5.26; it does not carry 3.5.27 (calls, Sumeet's own bot).
+
+- **`POST /api/browser/start`** (Agent Space's control token only, like every Browser app route). With no browser open, the bot's own default session opens on a start page (`https://www.google.com/`), with the bot's own agent-browser environment and socket folder (`AGENT_BROWSER_SOCKET_DIR`, else `~/.agent-browser`), exactly as its own `agent-browser open` would, so the bot carries on in that tab. A browser already open is answered as it is (the newest session), and two people opening the app at once start one browser. `GET /api/browser/control` says `start: true`; Agent Space's front door (223b978) asks only bots that say so.
+- **Tests:** `test-browser-control.js`, `test-web-browser-app.js`.
+
 ## 3.5.26
 
 The Browser app's phone and web view stays sharp after a page load. A canary build for the Browser canary bot first; the fleet waits for Sumeet's OK.

package/core/browser-control.js

@@ -17,6 +17,9 @@
 //     makes the bot's tab take its size, phone layout and touch, and then the driver's input goes
 //     straight to Chrome through that view (typed text too: input_text). Only with Chrome in its
 //     own container; elsewhere the browser keeps its size and input goes through agent-browser.
+//   - always on: a person who opens the Browser app while the bot's browser is not open gets it
+//     opened on a start page, in the bot's own default session (the one its `agent-browser open`
+//     uses), with the bot's own agent-browser environment, so the bot carries on in that tab.
 //   - hand back (release, a closed panel, or a lock found expired by the 30 s sweeper): the lock
 //     goes, the view and the socket close, any recording stops, and the bot is woken in the
 //     person's own chat (C6), but only when it waits for them: it sent them an Open Browser card,
@@ -26,6 +29,8 @@
 // Every route answers { status, body }; web.js writes it. Agent Space reaches these routes with
 // the control token only.
 
+const os = require("os");
+const path = require("path");
 const WebSocket = require("ws");
 const lockFile = require("./browser-lock");
 const view = require("./browser-view");
@@ -48,6 +53,10 @@ const ID_RE = /^[A-Za-z0-9._:-]{1,100}$/;
 // An address-bar command still running after this is killed, and the person told it failed.
 const NAVIGATE_TIMEOUT_MS = 20 * 1000;
 const NAVIGATIONS = new Set(["open", "back", "forward", "reload"]);
+// Always on: the page a browser opened for a person starts on, and the session it opens (the one
+// the bot's own `agent-browser` commands use when they name none).
+const START_PAGE = "https://www.google.com/";
+const START_SESSION = "default";
 
 // An Open Browser card counts as a request for a person this long.
 const REQUEST_MAX_AGE_MS = 24 * 60 * 60 * 1000;
@@ -521,7 +530,8 @@ async function control(body = {}) {
 }
 
 // navigate: true tells the front door this bot has the address bar (POST /api/browser/navigate);
-// fit: true that a take can bring the driver's view (core/browser-drive.js) and input_text works.
+// fit: true that a take can bring the driver's view (core/browser-drive.js) and input_text works;
+// start: true that it opens its browser on a start page when there is none (POST /api/browser/start).
 function status() {
   const lock = lockFile.liveLock(deps.now());
   return reply(200, {
@@ -530,9 +540,49 @@ function status() {
     recording: teach.activeSummary(),
     navigate: true,
     fit: drive.supported(),
+    start: true,
   });
 }
 
+// ---------------------------------------------------------------------------
+// Always on: POST /api/browser/start
+
+// The socket folder the bot's own agent-browser commands use (its agent runs without
+// XDG_RUNTIME_DIR): AGENT_BROWSER_SOCKET_DIR, else ~/.agent-browser.
+function agentSocketDir(env = process.env, home = os.homedir()) {
+  if (env.AGENT_BROWSER_SOCKET_DIR) return env.AGENT_BROWSER_SOCKET_DIR;
+  return home ? path.join(home, ".agent-browser") : null;
+}
+
+let starting = null;
+
+// { who: { userId } }. A browser already open is answered as it is (the newest session); otherwise
+// the bot's default session opens on the start page, as the bot's own `agent-browser open` would,
+// and two people opening the app at once start one browser. { session, started }.
+async function start({ who } = {}) {
+  if (!validWho(who)) return reply(400, { error: "bad-request" });
+  const open = deps.listSessions();
+  if (open.length) return reply(200, { session: open[0].name, started: false });
+  if (!starting) {
+    starting = (async () => {
+      const dir = agentSocketDir();
+      const run = await runAgentBrowser({ name: START_SESSION, dir }, ["open", START_PAGE], {
+        timeoutMs: deps.navigateTimeoutMs,
+        execFile: deps.execFile || undefined,
+      });
+      const listed = run.ok ? deps.listSessions().find((s) => s.name === START_SESSION) : null;
+      if (!listed) {
+        const why = run.timedOut ? `agent-browser took longer than ${Math.ceil(deps.navigateTimeoutMs / 1000)} s and was stopped` : run.ok ? "the browser did not come up" : "agent-browser failed";
+        console.error(`[browser] could not open the browser for ${who.userId}: ${why}`);
+        return reply(503, { error: "browser-unavailable" });
+      }
+      console.log(`[browser] opened the browser on the start page for ${who.userId}`);
+      return reply(200, { session: listed.name, started: true });
+    })().finally(() => { starting = null; });
+  }
+  return starting;
+}
+
 // ---------------------------------------------------------------------------
 // C3: input
 
@@ -738,7 +788,10 @@ module.exports = {
   LOCK_TTL_MS,
   MAX_EVENTS,
   NAVIGATE_TIMEOUT_MS,
+  START_PAGE,
+  START_SESSION,
   SWEEP_INTERVAL_MS,
+  agentSocketDir,
   checkEvent,
   chatOrigin,
   control,
@@ -748,6 +801,7 @@ module.exports = {
   navigate,
   release,
   renew,
+  start,
   startSweeper,
   status,
   sweepLock,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@inetafrica/open-claudia",
-  "version": "3.5.26",
+  "version": "3.5.28",
   "description": "An always-on, provider-agnostic coding-agent harness for Claude Code and OpenAI Codex via chat",
   "main": "bot.js",
   "bin": {

package/test-provider-language.js

@@ -617,7 +617,11 @@ assert.match(pkg.description, /provider-agnostic coding-agent harness/i);
 // only a waiting bot. A canary build for the Browser canary bot; the fleet waits for his OK.
 // 3.5.26: the same go-ahead; 3.5.25's pictures fell back to 1x after a page load on the
 // canary's Chrome 150, now nudged back to the view's density. Canary first, fleet on his OK.
-assert.strictEqual(pkg.version, "3.5.26", "release version must remain unchanged without explicit approval");
+// 3.5.28 approved under Sumeet's 2026-10-11 08:21 EAT go-ahead ("Yeah, that's fine") to the
+// resize fix and Chrome-style bar first, then the always-on browser with one restart of the
+// Browser canary bot: the Browser app opens the bot's browser when none is open. A canary build
+// for the Browser canary bot on 3.5.26; it does not carry 3.5.27 (calls, his own bot only).
+assert.strictEqual(pkg.version, "3.5.28", "release version must remain unchanged without explicit approval");
 for (const keyword of ["claude", "codex", "coding-agent", "provider-agnostic"]) {
   assert.ok(pkg.keywords.includes(keyword), `package keyword missing: ${keyword}`);
 }

package/web.js

@@ -382,9 +382,11 @@ const USER_ACCESS_ROUTES = new Set(["/api/auth/approve", "/api/auth/deny", "/api
 //   POST /api/browser/navigate                              the driver's address bar: open <url>
 //                                                           (public addresses only), back,
 //                                                           forward, reload
+//   POST /api/browser/start                                 always on: open the bot's browser on
+//                                                           a start page when none is open
 //   POST /api/browser/teach                                 Teach me: start / stop / cancel (C4)
 //   GET  /api/browser/teach[/<id>[/pictures/<n>]]?userId=   that person's recordings only
-const BROWSER_APP_ROUTE = /^\/api\/(?:apps\/check|browser\/(?:control|input|navigate|teach))(?:[/?]|$)/;
+const BROWSER_APP_ROUTE = /^\/api\/(?:apps\/check|browser\/(?:control|input|navigate|start|teach))(?:[/?]|$)/;
 const BROWSER_APP_MAX_BODY = 64 * 1024;
 const BROWSER_APP_USER_ID = /^[A-Za-z0-9._:-]{1,100}$/;
 
@@ -408,6 +410,7 @@ async function handleBrowserApp(req, res, body) {
   if (route === "/api/browser/control" && req.method === "GET") return answer(control.status());
   if (route === "/api/browser/input" && req.method === "POST") return answer(await control.input(parsed));
   if (route === "/api/browser/navigate" && req.method === "POST") return answer(await control.navigate(parsed));
+  if (route === "/api/browser/start" && req.method === "POST") return answer(await control.start(parsed));
   if (route === "/api/browser/teach" && req.method === "POST") return answer(await control.teachAction(parsed));
   if (req.method === "GET" && (route === "/api/browser/teach" || route.startsWith("/api/browser/teach/"))) {
     const teach = require("./core/browser-teach");