feat: let the agent show and drive the guest browser

app/src/main/java/com/yugahashimoto/andcode/feature/browser/BrowserCommandWatcher.kt

@@ -0,0 +1,45 @@
+package com.yugahashimoto.andcode.feature.browser
+
+import androidx.compose.runtime.Composable
+import androidx.compose.runtime.LaunchedEffect
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.delay
+import kotlinx.coroutines.withContext
+import org.json.JSONObject
+import java.io.File
+
+private const val COMMAND_FILE_RELATIVE_PATH = ".and-code/browser-command.json"
+private const val POLL_INTERVAL_MILLIS = 1000L
+
+/**
+ * Watches the active workspace for a browser command written by the in-guest agent
+ * (`.and-code/browser-command.json`, e.g. `{"action":"open","url":"http://127.0.0.1:8080/"}`)
+ * and opens the guest browser at the requested URL so the user can watch and join in.
+ */
+@Composable
+fun GuestBrowserCommandWatcher(
+    workspacePath: String?,
+    onOpenUrl: (String) -> Unit,
+) {
+    LaunchedEffect(workspacePath) {
+        val path = workspacePath ?: return@LaunchedEffect
+        val commandFile = File(path, COMMAND_FILE_RELATIVE_PATH)
+        while (true) {
+            delay(POLL_INTERVAL_MILLIS)
+            val url = withContext(Dispatchers.IO) { consumeOpenCommand(commandFile) }
+            if (url != null) {
+                onOpenUrl(url)
+            }
+        }
+    }
+}
+
+private fun consumeOpenCommand(commandFile: File): String? =
+    runCatching {
+        if (!commandFile.exists()) {
+            return null
+        }
+        val text = commandFile.readText()
+        commandFile.delete()
+        JSONObject(text).optString("url").takeIf { it.isNotBlank() }
+    }.getOrNull()

app/src/main/java/com/yugahashimoto/andcode/feature/browser/GuestBrowserScreen.kt

@@ -82,6 +82,10 @@ fun GuestBrowserScreen(
             AndroidView(
                 modifier = Modifier.weight(1f),
                 factory = { context ->
+                    // Lets the in-guest agent drive this WebView over CDP (the devtools abstract
+                    // socket is reachable from the guest, which shares the app's UID), so a page
+                    // can be shown to the user and operated by human and agent at the same time.
+                    WebView.setWebContentsDebuggingEnabled(true)
                     WebView(context).apply {
                         layoutParams =
                             ViewGroup.LayoutParams(

app/src/main/java/com/yugahashimoto/andcode/ui/AndCodeApp.kt

@@ -72,6 +72,7 @@ import com.yugahashimoto.andcode.feature.assistant.SpeechResult
 import com.yugahashimoto.andcode.feature.assistant.SpeechTranscriptAccumulator
 import com.yugahashimoto.andcode.feature.assistant.VoiceDictationOutcome
 import com.yugahashimoto.andcode.feature.assistant.VoiceDictationPolicy
+import com.yugahashimoto.andcode.feature.browser.GuestBrowserCommandWatcher
 import com.yugahashimoto.andcode.feature.chat.ChatHomeScreen
 import com.yugahashimoto.andcode.feature.chat.ChatViewModel
 import com.yugahashimoto.andcode.feature.chat.SubagentInfo
@@ -105,6 +106,7 @@ import com.yugahashimoto.andcode.ui.navigation.SCHEDULE_DETAIL_ROUTE_PATTERN
 import com.yugahashimoto.andcode.ui.navigation.SCHEDULE_EDIT_ARG_ID
 import com.yugahashimoto.andcode.ui.navigation.SCHEDULE_EDIT_ROUTE_PATTERN
 import com.yugahashimoto.andcode.ui.navigation.decodeRouteArg
+import com.yugahashimoto.andcode.ui.navigation.guestBrowserRoute
 import com.yugahashimoto.andcode.ui.navigation.scheduleDetailRoute
 import com.yugahashimoto.andcode.ui.navigation.scheduleEditRoute
 import com.yugahashimoto.andcode.ui.navigation.settingsNavGraph
@@ -563,6 +565,15 @@ fun AndCodeApp(
         }
     }
 
+    // Lets the in-guest agent pop the guest browser open for the user by dropping a command
+    // file into the active workspace (see GuestBrowserCommandWatcher).
+    GuestBrowserCommandWatcher(
+        workspacePath = chatState.selectedWorkspacePath,
+        onOpenUrl = { url ->
+            navController.navigate(guestBrowserRoute(url)) { launchSingleTop = true }
+        },
+    )
+
     val onHandoff: (String) -> Unit = { targetRuntimeId ->
         val prompt = buildHandoffPrompt(chatState.messages)
         pendingHandoffPrompt = targetRuntimeId to prompt

app/src/main/java/com/yugahashimoto/andcode/ui/navigation/Routes.kt

@@ -44,6 +44,10 @@ const val SCHEDULE_EDIT_ROUTE_PATTERN = "$ROUTE_SCHEDULE_EDIT?$SCHEDULE_EDIT_ARG
 const val ROUTE_CODE_VIEWER = "code-viewer"
 const val ROUTE_TERMINAL = "terminal"
 const val ROUTE_GUEST_BROWSER = "guest-browser"
+const val GUEST_BROWSER_ARG_URL = "url"
+const val GUEST_BROWSER_ROUTE_PATTERN = "$ROUTE_GUEST_BROWSER?$GUEST_BROWSER_ARG_URL={$GUEST_BROWSER_ARG_URL}"
+
+fun guestBrowserRoute(url: String): String = "$ROUTE_GUEST_BROWSER?$GUEST_BROWSER_ARG_URL=${encodeRouteArg(url)}"
 
 const val CODE_VIEWER_ROUTE_PATTERN = "$ROUTE_CODE_VIEWER/{runtimeId}/{workspacePath}/{filePath}"
 

app/src/main/java/com/yugahashimoto/andcode/ui/navigation/WorkspaceNavGraph.kt

@@ -185,9 +185,12 @@ fun NavGraphBuilder.workspaceNavGraph(
         )
     }
 
-    composable(ROUTE_GUEST_BROWSER) {
+    composable(GUEST_BROWSER_ROUTE_PATTERN) { backStack ->
+        val requestedUrl = backStack.arguments?.getString(GUEST_BROWSER_ARG_URL)?.let { decodeRouteArg(it) }
         GuestBrowserScreen(
-            initialUrl = app.localRuntimeManager.installedPort()?.let { "http://127.0.0.1:$it/" }.orEmpty(),
+            initialUrl =
+                requestedUrl
+                    ?: app.localRuntimeManager.installedPort()?.let { "http://127.0.0.1:$it/" }.orEmpty(),
             onBack = { navController.popBackStack() },
         )
     }