feat(agent): Improve PhoneAgent stability and add research docs

- Enhance PhoneAgent with better error handling and retry logic
- Refactor input handling (KeyboardHelper, TextInputManager)
- Clean up unused code across multiple modules
- Add multi-model agent architecture research documentation
- Bump version to 0.0.3

Author: Luokavin

.gitignore

@@ -55,5 +55,8 @@ keystore_base64.txt
 # Logcat dumps
 logcat*
 
+# AutoGLM logs
+autoglm_logs_*/
+
 # Dev profiles config
 dev_profiles.json

app/build.gradle.kts

@@ -12,8 +12,8 @@ android {
         applicationId = "com.kevinluo.autoglm"
         minSdk = 24
         targetSdk = 34
-        versionCode = 2
-        versionName = "0.0.2"
+        versionCode = 3
+        versionName = "0.0.3"
 
         testInstrumentationRunner = "androidx.test.runner.AndroidJUnitRunner"
     }

app/src/main/java/com/kevinluo/autoglm/ComponentManager.kt

@@ -17,6 +17,7 @@ import com.kevinluo.autoglm.screenshot.ScreenshotService
 import com.kevinluo.autoglm.settings.SettingsManager
 import com.kevinluo.autoglm.ui.FloatingWindowService
 import com.kevinluo.autoglm.util.HumanizedSwipeGenerator
+import com.kevinluo.autoglm.util.Logger
 
 /**
  * Centralized component manager for dependency injection and lifecycle management.
@@ -27,7 +28,6 @@ import com.kevinluo.autoglm.util.HumanizedSwipeGenerator
  * - Lifecycle-aware component management
  * - Clean separation of concerns
  * 
- * Requirements: All (integration)
  */
 class ComponentManager private constructor(private val context: Context) {
 
@@ -165,7 +165,7 @@ class ComponentManager private constructor(private val context: Context) {
      * @param service The connected UserService
      */
     fun onServiceConnected(service: IUserService) {
-        android.util.Log.i(TAG, "UserService connected, initializing components")
+        Logger.i(TAG, "UserService connected, initializing components")
         userService = service
         initializeServiceDependentComponents()
     }
@@ -175,7 +175,7 @@ class ComponentManager private constructor(private val context: Context) {
      * Cleans up service-dependent components.
      */
     fun onServiceDisconnected() {
-        android.util.Log.i(TAG, "UserService disconnected, cleaning up components")
+        Logger.i(TAG, "UserService disconnected, cleaning up components")
         userService = null
         cleanupServiceDependentComponents()
     }
@@ -215,7 +215,7 @@ class ComponentManager private constructor(private val context: Context) {
             historyManager = historyManager
         )
 
-        android.util.Log.i(TAG, "All service-dependent components initialized")
+        Logger.i(TAG, "All service-dependent components initialized")
     }
 
     /**
@@ -229,7 +229,7 @@ class ComponentManager private constructor(private val context: Context) {
         _textInputManager = null
         _deviceExecutor = null
 
-        android.util.Log.i(TAG, "Service-dependent components cleaned up")
+        Logger.i(TAG, "Service-dependent components cleaned up")
     }
 
     /**
@@ -238,7 +238,7 @@ class ComponentManager private constructor(private val context: Context) {
      */
     fun reinitializeAgent() {
         if (userService == null) {
-            android.util.Log.w(TAG, "Cannot reinitialize agent: UserService not connected")
+            Logger.w(TAG, "Cannot reinitialize agent: UserService not connected")
             return
         }
 
@@ -258,7 +258,7 @@ class ComponentManager private constructor(private val context: Context) {
             historyManager = historyManager
         )
 
-        android.util.Log.i(TAG, "PhoneAgent reinitialized with new configuration")
+        Logger.i(TAG, "PhoneAgent reinitialized with new configuration")
     }
 
     /**
@@ -295,7 +295,7 @@ class ComponentManager private constructor(private val context: Context) {
      * Should be called when the application is being destroyed.
      */
     fun cleanup() {
-        android.util.Log.i(TAG, "Cleaning up all components")
+        Logger.i(TAG, "Cleaning up all components")
         cleanupServiceDependentComponents()
         _modelClient = null
         _appResolver = null

app/src/main/java/com/kevinluo/autoglm/MainActivity.kt

@@ -50,7 +50,6 @@ import rikka.shizuku.Shizuku
  * The activity implements [PhoneAgentListener] to receive callbacks
  * during task execution for UI updates.
  *
- * Requirements: 1.1, 2.1, 2.2
  */
 class MainActivity : AppCompatActivity(), PhoneAgentListener {
 
@@ -390,7 +389,7 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
             openShizukuApp()
         }
 
-        // Settings button - Requirements: 6.1
+        // Settings button
         settingsBtn.setOnClickListener {
             startActivity(Intent(this, SettingsActivity::class.java))
         }
@@ -416,12 +415,12 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
             com.kevinluo.autoglm.input.KeyboardHelper.openInputMethodSettings(this)
         }
 
-        // Start task button - Requirements: 1.1
+        // Start task button
         startTaskBtn.setOnClickListener {
             startTask()
         }
 
-        // Cancel task button - Requirements: 1.4
+        // Cancel task button
         cancelTaskBtn.setOnClickListener {
             cancelTask()
         }
@@ -604,7 +603,6 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
      * Called after UserService is connected. Sets up the agent listener
      * and confirmation callback for sensitive operations.
      *
-     * Requirements: 1.1, 2.1, 2.2
      */
     private fun initializePhoneAgent() {
         if (!componentManager.isServiceConnected) {
@@ -667,7 +665,6 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
      * Validates the task description, checks agent state, starts the
      * floating window service, and launches the task in a coroutine.
      *
-     * Requirements: 1.1, 2.1, 2.2
      */
     private fun startTask() {
         val taskDescription = taskInput.text?.toString()?.trim() ?: ""
@@ -779,7 +776,6 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
      * Cancels the agent, resets its state, and updates the UI
      * to reflect the cancelled status.
      *
-     * Requirements: 1.1, 2.1, 2.2
      */
     private fun cancelTask() {
         Logger.i(TAG, "Cancelling task")
@@ -865,7 +861,6 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
      *
      * @param status The new task status to display
      *
-     * Requirements: 1.1, 2.1, 2.2
      */
     private fun updateTaskStatus(status: TaskStatus) {
         val (text, colorRes) = when (status) {
@@ -898,7 +893,6 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
      *
      * @param stepNumber The current step number
      *
-     * Requirements: 1.1, 2.1, 2.2
      */
     override fun onStepStarted(stepNumber: Int) {
         runOnUiThread {
@@ -916,7 +910,6 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
      *
      * @param thinking The model's thinking text
      *
-     * Requirements: 1.1, 2.1, 2.2
      */
     override fun onThinkingUpdate(thinking: String) {
         runOnUiThread {
@@ -931,7 +924,6 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
      *
      * @param action The action that was executed
      *
-     * Requirements: 1.1, 2.1, 2.2
      */
     override fun onActionExecuted(action: AgentAction) {
         runOnUiThread {
@@ -948,7 +940,6 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
      *
      * @param message The completion message
      *
-     * Requirements: 1.1, 2.1, 2.2
      */
     override fun onTaskCompleted(message: String) {
         runOnUiThread {
@@ -967,7 +958,6 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
      *
      * @param error The error message
      *
-     * Requirements: 1.1, 2.1, 2.2
      */
     override fun onTaskFailed(error: String) {
         runOnUiThread {
@@ -983,7 +973,6 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
      *
      * Note: Floating window hide is handled by ScreenshotService.
      *
-     * Requirements: 1.1, 2.1, 2.2
      */
     override fun onScreenshotStarted() {
         // Floating window hide is handled by ScreenshotService
@@ -994,7 +983,6 @@ class MainActivity : AppCompatActivity(), PhoneAgentListener {
      *
      * Note: Floating window show is handled by ScreenshotService.
      *
-     * Requirements: 1.1, 2.1, 2.2
      */
     override fun onScreenshotCompleted() {
         // Floating window show is handled by ScreenshotService

app/src/main/java/com/kevinluo/autoglm/action/ActionHandler.kt

@@ -24,7 +24,6 @@ import kotlinx.coroutines.delay
  * @param textInputManager Manager for text input operations
  * @param floatingWindowProvider Optional provider for floating window controller
  *
- * Requirements: 1.1, 1.7, 2.1, 2.2
  */
 class ActionHandler(
     private val deviceExecutor: DeviceExecutor,
@@ -109,7 +108,6 @@ class ActionHandler(
      * @param screenHeight Current screen height in pixels
      * @return The result of the action execution
      *
-     * Requirements: 1.7
      */
     suspend fun execute(
         action: AgentAction,
@@ -160,7 +158,6 @@ class ActionHandler(
      * Executes a Tap action.
      * Hides floating window before tap to prevent touch interception.
      * Uses try-finally to ensure floating window is restored even if tap fails.
-     * Requirements: 5.1, 5.2
      */
     private suspend fun executeTap(
         action: AgentAction.Tap,
@@ -200,7 +197,6 @@ class ActionHandler(
      * Executes a Swipe action.
      * Hides floating window before swipe to prevent touch interception.
      * Uses try-finally to ensure floating window is restored even if swipe fails.
-     * Requirements: 5.3, 5.4, 10.1-10.5
      */
     private suspend fun executeSwipe(
         action: AgentAction.Swipe,
@@ -255,7 +251,6 @@ class ActionHandler(
      * 3. The BroadcastReceiver is registered
      * 
      * Uses try-finally to ensure floating window is restored even if typing fails.
-     * Requirements: 5.5
      */
     private suspend fun executeType(action: AgentAction.Type): ActionResult {
         // Hide floating window to ensure target app has focus
@@ -276,7 +271,6 @@ class ActionHandler(
     /**
      * Executes a TypeName action (same as Type).
      * Uses try-finally to ensure floating window is restored even if typing fails.
-     * Requirements: 5.6
      */
     private suspend fun executeTypeName(action: AgentAction.TypeName): ActionResult {
         // Hide floating window to ensure target app has focus
@@ -300,7 +294,6 @@ class ActionHandler(
      * If not found, returns success with a message instructing the model
      * to find the app icon on screen (home screen or app drawer).
      * 
-     * Requirements: 5.7, 5.8
      */
     private suspend fun executeLaunch(action: AgentAction.Launch): ActionResult {
         Logger.d(TAG, "Launching app: ${action.app}")
@@ -381,7 +374,6 @@ class ActionHandler(
      * Executes a Back action.
      * First dismisses the soft keyboard (if shown) to ensure the back action
      * actually navigates back instead of just closing the keyboard.
-     * Requirements: 5.9
      */
     private suspend fun executeBack(): ActionResult {
         // First, dismiss keyboard with ESCAPE key to ensure Back actually navigates
@@ -401,7 +393,6 @@ class ActionHandler(
 
     /**
      * Executes a Home action.
-     * Requirements: 5.10
      */
     private suspend fun executeHome(): ActionResult {
         val result = deviceExecutor.pressKey(DeviceExecutor.KEYCODE_HOME)
@@ -415,7 +406,6 @@ class ActionHandler(
 
     /**
      * Executes a VolumeUp action.
-     * Requirements: 5.11
      */
     private suspend fun executeVolumeUp(): ActionResult {
         val result = deviceExecutor.pressKey(DeviceExecutor.KEYCODE_VOLUME_UP)
@@ -429,7 +419,6 @@ class ActionHandler(
 
     /**
      * Executes a VolumeDown action.
-     * Requirements: 5.12
      */
     private suspend fun executeVolumeDown(): ActionResult {
         val result = deviceExecutor.pressKey(DeviceExecutor.KEYCODE_VOLUME_DOWN)
@@ -443,7 +432,6 @@ class ActionHandler(
 
     /**
      * Executes a Power action.
-     * Requirements: 5.13
      */
     private suspend fun executePower(): ActionResult {
         val result = deviceExecutor.pressKey(DeviceExecutor.KEYCODE_POWER)
@@ -460,7 +448,6 @@ class ActionHandler(
      * Executes a LongPress action.
      * Hides floating window before long press to prevent touch interception.
      * Uses try-finally to ensure floating window is restored even if long press fails.
-     * Requirements: 5.14
      */
     private suspend fun executeLongPress(
         action: AgentAction.LongPress,
@@ -497,7 +484,6 @@ class ActionHandler(
      * Executes a DoubleTap action.
      * Hides floating window before double tap to prevent touch interception.
      * Uses try-finally to ensure floating window is restored even if double tap fails.
-     * Requirements: 5.15
      */
     private suspend fun executeDoubleTap(
         action: AgentAction.DoubleTap,
@@ -527,7 +513,6 @@ class ActionHandler(
 
     /**
      * Executes a Wait action.
-     * Requirements: 5.16
      */
     private suspend fun executeWait(action: AgentAction.Wait): ActionResult {
         val durationMs = (action.durationSeconds * 1000).toLong()
@@ -537,7 +522,6 @@ class ActionHandler(
 
     /**
      * Executes a TakeOver action.
-     * Requirements: 5.17
      */
     private suspend fun executeTakeOver(action: AgentAction.TakeOver): ActionResult {
         confirmationCallback?.onTakeOverRequested(action.message)
@@ -546,7 +530,6 @@ class ActionHandler(
 
     /**
      * Executes an Interact action.
-     * Requirements: 5.18
      */
     private suspend fun executeInteract(action: AgentAction.Interact): ActionResult {
         val selectedIndex = confirmationCallback?.onInteractionRequired(action.options) ?: -1
@@ -560,7 +543,6 @@ class ActionHandler(
 
     /**
      * Executes a Note action.
-     * Requirements: 5.19
      */
     private suspend fun executeNote(action: AgentAction.Note): ActionResult {
         // Note action just records the message, no device operation needed
@@ -569,7 +551,6 @@ class ActionHandler(
 
     /**
      * Executes a CallApi action.
-     * Requirements: 5.20
      */
     private suspend fun executeCallApi(action: AgentAction.CallApi): ActionResult {
         // CallApi action is handled by the agent layer, not device operations
@@ -578,7 +559,6 @@ class ActionHandler(
 
     /**
      * Executes a Finish action.
-     * Requirements: 5.21
      */
     private suspend fun executeFinish(action: AgentAction.Finish): ActionResult {
         return ActionResult(true, true, action.message)

app/src/main/java/com/kevinluo/autoglm/action/ActionParser.kt

@@ -13,7 +13,6 @@ import com.kevinluo.autoglm.util.Logger
  * and system key actions. It validates coordinate ranges and provides detailed
  * error messages for parsing failures.
  *
- * Requirements: 1.1, 2.1, 2.2
  */
 object ActionParser {
 

app/src/main/java/com/kevinluo/autoglm/action/AgentAction.kt

@@ -10,7 +10,6 @@ package com.kevinluo.autoglm.action
  * The sealed class pattern ensures exhaustive handling of all action types
  * and provides type-safe action processing.
  *
- * Requirements: 2.1, 7.2
  */
 sealed class AgentAction {