docs: update Android capability status

README.md

@@ -12,9 +12,9 @@ git@git.nexlab.net:lisa/hermes-node-android.git
 
 ## Current status
 
-Buildable first test APK.
+Working Android node APK that connects to the Hermes Node Gateway and implements a meaningful first real capability set.
 
-Implemented enough to install on a phone and connect it to the Hermes Node Gateway:
+Implemented:
 
 - Android/Kotlin app project with Gradle wrapper.
 - Config UI for gateway URL, token, node name, TLS mode, and advertised capabilities.
@@ -23,29 +23,46 @@ Implemented enough to install on a phone and connect it to the Hermes Node Gatew
 - Reconnect loop and heartbeat frames.
 - App-local config via `SharedPreferences`; no hardcoded token.
 - Status display in the app.
+- Permission/setup helpers in the UI for camera, microphone, Accessibility, and screenshot consent.
 - Debug APK build verified locally.
 
-Baseline capability handlers:
+Implemented capability handlers:
 
 - `computer_control`
   - `launch_app`: implemented via package launch intent.
   - `open_url`: implemented via `ACTION_VIEW`.
   - `open_intent`: implemented for implicit Android intents.
-  - gesture/key/text actions return explicit unsupported until AccessibilityService backend is added.
+  - `tap`, `swipe`, `long_press`: implemented via AccessibilityService gestures.
+  - `type_text`: implemented via Accessibility focused-node text injection.
+  - `key_press`: implemented for Android global actions like `back`, `home`, `recents`, `notifications`, `quick_settings`.
 - `desktop_observe`
   - `screen_info`: implemented.
+  - `active_window`: implemented when AccessibilityService is enabled.
+  - `clipboard_get`: implemented.
   - cursor/mouse-position compatibility returns Android-specific note.
-  - screenshots/active-window/clipboard return explicit unsupported until Android permission backends are added.
+  - `list_windows` and `window_geometry`: compatibility implementations.
+  - screenshot consent flow is wired, but bitmap capture/export is still incomplete.
 - `audio_control`
   - `get_audio_status`: implemented.
-  - other actions return explicit unsupported until media/mic backend is added.
+  - `list_audio_devices`: implemented via `AudioDeviceInfo`.
+  - `capture_input`: implemented via `AudioRecord`, returning WAV as base64.
+  - `capture_output`: explicitly unsupported on stock Android.
 - `camera_control`
   - `list_cameras`: implemented via Camera2.
   - `get_camera_status`: implemented.
-  - capture/injection actions return explicit unsupported until CameraX/MediaProjection UX exists.
+  - `capture_frame`: implemented via Camera2 + `ImageReader`, returning base64 JPEG.
+  - `inject_video`: explicitly unsupported on stock Android.
 - `exec`
   - advertised only if enabled in UI, but backend intentionally returns unsupported for now.
 
+Still incomplete by platform reality or next-stage engineering:
+
+- `desktop_observe.screenshot` / `region_screenshot`: MediaProjection consent is wired, but the capture pipeline still needs completion.
+- `audio_control.play_audio`: needs cache/download and playback backend.
+- `camera_control.capture_video`: needs longer-lived visible recording pipeline and output policy.
+- `browser_control`: not implemented yet.
+- `exec`: intentionally disabled as non-parity Android shell execution.
+
 ## Capability naming decision
 
 Use existing gateway capability names where practical:
@@ -102,11 +119,18 @@ app/build/outputs/apk/debug/app-debug.apk
 4. Tap **Save & start node**.
 5. Android should show a persistent foreground-service notification.
 6. On Hermes, check `node_list` / gateway node status for the Android node name.
-7. Smoke-test safe actions:
+7. In the Android app, also tap:
+   - **Grant camera/microphone permissions**
+   - **Open Accessibility settings** and enable Hermes Node if you want gestures/active window support
+   - **Grant screenshot consent** if you want to prepare for screenshot testing
+8. Smoke-test actions from Hermes:
    - `desktop_observe` → `screen_info`
+   - `desktop_observe` → `active_window`
    - `audio_control` → `get_audio_status`
+   - `audio_control` → `list_audio_devices`
    - `camera_control` → `list_cameras`
    - `computer_control` → `open_url`
+   - `computer_control` → `tap` / `swipe` after Accessibility is enabled
 
 ## Repository layout
 

docs/ANDROID_CAPABILITIES.md

@@ -8,15 +8,21 @@ The Android agent should expose **the same gateway capability names** as Linux w
 
 Meaning on Android: phone/device control.
 
-Current first-test APK actions:
+Current implemented actions:
 
 - `launch_app`: implemented; starts app by package name.
 - `open_url`: implemented; opens URL via Android intent.
 - `open_intent`: implemented for implicit Android intent bridge.
-- `key_press`: returns explicit unsupported until AccessibilityService/privileged input backend exists.
-- `tap`, `swipe`, `long_press`: return explicit unsupported until AccessibilityService gesture dispatch exists.
-- `type_text`: returns explicit unsupported until accessibility/IME-backed text entry exists.
-- `get_active_window`: route through `desktop_observe`; foreground app/activity needs UsageStats or Accessibility grant.
+- `key_press`: implemented for Android global actions via AccessibilityService (`back`, `home`, `recents`, `notifications`, `quick_settings`, `power_dialog`).
+- `tap`, `swipe`, `long_press`: implemented via AccessibilityService gesture dispatch.
+- `type_text`: implemented via Accessibility node focused-text injection.
+- `mouse_move`: implemented as Android no-op compatibility response; Android has no persistent cursor.
+- `get_active_window`: still routed through `desktop_observe`.
+
+Explicit limitations that remain honest:
+
+- Arbitrary raw key injection like desktop Enter/volume semantics is not available through normal unprivileged Android app APIs.
+- Anything requiring shell-level input injection or hidden APIs is intentionally not faked.
 
 Keep the gateway name `computer_control` unless gateway-side UX later needs a friendly alias. Compatibility beats naming purity here.
 
@@ -24,33 +30,35 @@ Keep the gateway name `computer_control` unless gateway-side UX later needs a fr
 
 Meaning on Android: screen/device observation.
 
-Current first-test APK actions:
+Current implemented actions:
 
 - `screen_info`: implemented; display metrics, density, orientation, interactive state.
-- `active_window` / `get_active_window`: explicit unsupported until UsageStats or AccessibilityService grant flow exists.
-- `screenshot` / `region_screenshot`: explicit unsupported until MediaProjection consent flow exists.
-- `clipboard_get`: explicit unsupported until Android foreground/API restrictions are handled.
+- `active_window` / `get_active_window`: implemented via AccessibilityService when enabled.
+- `clipboard_get`: implemented.
 - `cursor_position` / `mouse_position`: implemented as compatibility shim; Android has touch focus, not a persistent cursor.
+- `list_windows`: implemented as Android foreground-window model compatibility.
+- `window_geometry`: implemented as screen geometry fallback.
+- `screenshot` / `region_screenshot`: consent flow is wired, but bitmap capture/export is still incomplete until the MediaProjection virtual-display pipeline is finished.
 
 ### `audio_control`
 
-Current first-test APK actions:
+Current implemented actions:
 
 - `get_audio_status`: implemented; stream volume, ringer mode, speakerphone flag, mic mute flag.
-- `list_audio_devices`: explicit unsupported until AudioDeviceInfo enumeration is wired.
-- `play_audio`: explicit unsupported until download/cache + MediaPlayer/ExoPlayer backend exists.
-- `capture_input`: explicit unsupported until microphone permission and foreground recording UX exist.
-- `capture_output`: explicit unsupported; normal Android apps generally cannot capture arbitrary output audio.
+- `list_audio_devices`: implemented via `AudioDeviceInfo`.
+- `capture_input`: implemented via `AudioRecord`, returning WAV payload as base64 when microphone permission is granted.
+- `play_audio`: still incomplete until download/cache + MediaPlayer/ExoPlayer backend exists.
+- `capture_output`: explicitly unsupported; normal Android apps generally cannot capture arbitrary output audio.
 
 ### `camera_control`
 
-Current first-test APK actions:
+Current implemented actions:
 
-- `list_cameras`: implemented with Camera2 camera inventory.
+- `list_cameras`: implemented with Camera2 camera inventory and JPEG size hints.
 - `get_camera_status`: implemented with backend/status metadata.
-- `capture_frame`: explicit unsupported until CameraX/Camera2 capture session and permission UX exist.
-- `capture_video`: explicit unsupported until CameraX/Camera2 capture session and visible UX exist.
-- `inject_video`: explicit unsupported on normal Android; no fake parity.
+- `capture_frame`: implemented with Camera2 + `ImageReader`, returning base64 JPEG when camera permission is granted.
+- `capture_video`: still incomplete; this needs a longer-lived visible recording pipeline and storage/streaming policy.
+- `inject_video`: explicitly unsupported on normal Android; no fake parity.
 
 ### `browser_control`