Skip to content

armgit / cua

Public Imported from GitHub
0

Scale computer-use 2.0 with open-source drivers, cross-OS sandboxes, and benchmarks for training, evaluation, and data generation.

Add nested virtualization to cuabot, add `agent-browser` and `agent-device` to Dockerfile (#1022)

ddupont964127337606
diff --git a/README.md b/README.md
index d8ae95a0cc..43e5fbd96e 100644
--- a/README.md
+++ b/README.md
@@ -40,6 +40,8 @@ cuabot --type "hello"
cuabot --click <x> <y> [button]
```
+Built-in support for `agent-browser` and `agent-device` (iOS, Android) out of the box.
+
<div align="center">
**[Get Started](https://cua.ai/docs/cuabot/cuabot)** | **[Installation](https://cua.ai/docs/cuabot/install)** | *First spotted at [ClawCon](https://www.claw-con.com/)*
diff --git a/libs/cuabot/Dockerfile b/libs/cuabot/Dockerfile
index b264ff622b..2bc919c84e 100644
--- a/libs/cuabot/Dockerfile
+++ b/libs/cuabot/Dockerfile
@@ -19,7 +19,8 @@ RUN apt-get update && apt-get install -y \
python3-pip \
lsb-release \
netcat-openbsd \
- git
+ git \
+ dos2unix
# Install xpra from official repo (must be v6+)
RUN apt-get update \
@@ -31,10 +32,10 @@ RUN apt-get update \
&& XPRA_VERSION=$(xpra --version | grep -oP 'v\K[0-9]+' | head -1) \
&& if [ "$XPRA_VERSION" -lt 5 ]; then echo "ERROR: xpra version must be >= 5, got v$XPRA_VERSION"; exit 1; fi
-# Install Node.js 20.x
+# Install Node.js 22.x (agent-device requires >=22)
RUN mkdir -p /etc/apt/keyrings \
&& curl -fsSL https://deb.nodesource.com/gpgkey/nodesource-repo.gpg.key | gpg --dearmor | tee /etc/apt/keyrings/nodesource.gpg > /dev/null \
- && echo "deb [signed-by=/etc/apt/keyrings/nodesource.gpg] https://deb.nodesource.com/node_20.x nodistro main" | tee /etc/apt/sources.list.d/nodesource.list \
+ && echo "deb [signed-by=/etc/apt/keyrings/nodesource.gpg] https://deb.nodesource.com/node_22.x nodistro main" | tee /etc/apt/sources.list.d/nodesource.list \
&& apt-get update && apt-get install -y nodejs
# Install remaining packages
@@ -48,6 +49,48 @@ RUN apt-get install -y \
&& DEBIAN_FRONTEND=noninteractive npx playwright install-deps \
&& apt-get clean && rm -rf /var/lib/apt/lists/*
+# Install Android SDK dependencies for agent-device
+RUN apt-get update && apt-get install -y --no-install-recommends \
+ openjdk-17-jdk-headless \
+ unzip \
+ libpulse0 \
+ libgl1-mesa-glx \
+ libnss3 \
+ libxcomposite1 \
+ libxcursor1 \
+ libxi6 \
+ libxtst6 \
+ libxrandr2 \
+ libasound2 \
+ wmctrl \
+ && apt-get clean && rm -rf /var/lib/apt/lists/*
+
+# Install agent-device globally
+RUN npm install -g agent-device
+
+# Set Android SDK environment variables
+ENV ANDROID_SDK_ROOT=/home/user/android-sdk
+ENV JAVA_HOME=/usr/lib/jvm/java-17-openjdk-amd64
+ENV PATH="${ANDROID_SDK_ROOT}/cmdline-tools/latest/bin:${ANDROID_SDK_ROOT}/platform-tools:${ANDROID_SDK_ROOT}/emulator:${PATH}"
+
+# Install Android SDK command-line tools
+RUN mkdir -p "${ANDROID_SDK_ROOT}/cmdline-tools" \
+ && cd /tmp \
+ && wget -q "https://dl.google.com/android/repository/commandlinetools-linux-11076708_latest.zip" -O cmdline-tools.zip \
+ && unzip -q cmdline-tools.zip -d "${ANDROID_SDK_ROOT}/cmdline-tools" \
+ && mv "${ANDROID_SDK_ROOT}/cmdline-tools/cmdline-tools" "${ANDROID_SDK_ROOT}/cmdline-tools/latest" \
+ && rm cmdline-tools.zip \
+ && chown -R user:user "${ANDROID_SDK_ROOT}"
+
+# Install platform-tools, emulator, and platforms (NOT the system image - too large)
+RUN yes | sdkmanager --licenses 2>/dev/null || true \
+ && sdkmanager "platform-tools" "emulator" "platforms;android-35" \
+ && chown -R user:user "${ANDROID_SDK_ROOT}"
+
+# Symlink adb/emulator into PATH
+RUN ln -sf /home/user/android-sdk/platform-tools/adb /usr/local/bin/adb \
+ && ln -sf /home/user/android-sdk/emulator/emulator /usr/local/bin/emulator
+
# Install Chromium from xtradeb PPA (avoids snap requirement on Ubuntu 22.04)
RUN add-apt-repository ppa:xtradeb/apps -y \
&& apt-get update \
@@ -82,8 +125,6 @@ RUN agent-browser install
# Set Chromium as default browser (needed for some agent CLI onboardings)
RUN xdg-settings set default-web-browser chromium.desktop 2>/dev/null || true
-## ==== preinstalled apps for the agent ====
-
# Python plotting libraries
RUN pip3 install --no-cache-dir matplotlib numpy pandas seaborn plotly
@@ -106,6 +147,14 @@ RUN mkdir -p ~/.local/bin && \
printf '#!/bin/bash\nnpx @google/gemini-cli "$@"' > ~/.local/bin/gemini && \
chmod +x ~/.local/bin/codex ~/.local/bin/aider ~/.local/bin/openclaw ~/.local/bin/vibe ~/.local/bin/gemini
+# Wrapper for agent-device (handles install command for system image download)
+COPY --chown=user:user src/scripts/agent-device-wrapper.sh /home/user/.local/bin/agent-device
+RUN dos2unix /home/user/.local/bin/agent-device && chmod +x /home/user/.local/bin/agent-device
+
+# Copy entrypoint script
+COPY --chown=user:user src/scripts/entrypoint.sh /home/user/.local/bin/entrypoint.sh
+RUN dos2unix /home/user/.local/bin/entrypoint.sh && chmod +x /home/user/.local/bin/entrypoint.sh
+
ENV DISPLAY=:100
ENV XDG_RUNTIME_DIR=/tmp/runtime-user
ENV AGENT_BROWSER_HEADED=1
@@ -122,4 +171,4 @@ EXPOSE 10000
USER user
# Note: dbus-daemon runs as session bus (not system) to avoid needing root
-CMD ["sh", "-c", "xpra --version && dbus-daemon --session --fork --address=unix:path=/tmp/dbus-session 2>/dev/null || true && xpra seamless :100 --sharing=yes --no-daemon --bind-tcp=0.0.0.0:10000 --html=on --dpi=96"]
+CMD ["/home/user/.local/bin/entrypoint.sh"]
diff --git a/libs/cuabot/src/cuabotd.ts b/libs/cuabot/src/cuabotd.ts
index c81fd5d6ee..f09b3f4bb4 100644
--- a/libs/cuabot/src/cuabotd.ts
+++ b/libs/cuabot/src/cuabotd.ts
@@ -34,6 +34,7 @@ const execAsync = promisify(exec);
const DEFAULT_PORT = 7842;
const CONFIG_DIR = join(homedir(), ".cuabot");
const CLAUDE_CONFIG_DIR = join(CONFIG_DIR, "user", ".claude");
+const ANDROID_SDK_IMAGES_DIR = join(CONFIG_DIR, "user", "android-sdk", "system-images");
const IMAGE_NAME = "trycua/cuabot:latest";
const BASE_CONTAINER_NAME = "cuabot-xpra";
@@ -49,6 +50,7 @@ export function getSessionName(): string | null {
return sessionName;
}
+
// Find an available HTTP server port
async function findAvailableServerPort(preferredPort: number): Promise<number> {
const net = await import("net");
@@ -121,6 +123,10 @@ function ensureConfigDir() {
if (!existsSync(CLAUDE_CONFIG_DIR)) {
mkdirSync(CLAUDE_CONFIG_DIR, { recursive: true });
}
+ // Ensure Android SDK system-images directory exists for volume mount
+ if (!existsSync(ANDROID_SDK_IMAGES_DIR)) {
+ mkdirSync(ANDROID_SDK_IMAGES_DIR, { recursive: true });
+ }
}
// Docker helpers
@@ -214,6 +220,7 @@ async function ensureContainer(): Promise<number> {
// Convert Windows path to Docker-compatible path for volume mount
const toDockerPath = (p: string) => p.replace(/\\/g, "/").replace(/^([A-Z]):/, (_, letter) => `/${letter.toLowerCase()}`);
const claudeConfigPath = toDockerPath(CLAUDE_CONFIG_DIR);
+ const androidSdkImagesPath = toDockerPath(ANDROID_SDK_IMAGES_DIR);
const telemetryEnabled = getTelemetryEnabled();
const cuabotName = sessionName || "cuabot";
@@ -230,9 +237,17 @@ async function ensureContainer(): Promise<number> {
"-e", `CUABOT_COLOR=${cuabotColor}`,
"--add-host=host.docker.internal:host-gateway",
"-v", `${claudeConfigPath}:/home/user/.claude`,
- IMAGE_NAME,
+ "-v", `${androidSdkImagesPath}:/home/user/android-sdk/system-images`,
];
+ // Add KVM device passthrough for nested virtualization
+ if (process.env.CUABOT_NESTED_VIRT === "1") {
+ createCmd.push("--device=/dev/kvm");
+ console.log("[server] Nested virtualization enabled (--device=/dev/kvm)");
+ }
+
+ createCmd.push(IMAGE_NAME);
+
await execAsync(createCmd.join(" "));
await execAsync(`docker start ${getContainerName()}`);
@@ -1160,6 +1175,16 @@ export async function stopServer(): Promise<boolean> {
} catch { }
}
+ // Stop and remove Docker container
+ try {
+ await execAsync(`docker stop ${getContainerName()}`);
+ console.log(`Stopped container ${getContainerName()}`);
+ } catch { }
+ try {
+ await execAsync(`docker rm ${getContainerName()}`);
+ console.log(`Removed container ${getContainerName()}`);
+ } catch { }
+
// Clean up files
try { unlinkSync(getPidFile()); } catch { }
try { unlinkSync(getPortFile()); } catch { }
diff --git a/libs/cuabot/src/prompts/SYSTEM.md b/libs/cuabot/src/prompts/SYSTEM.md
index 74645e151d..676c25c5ef 100644
--- a/libs/cuabot/src/prompts/SYSTEM.md
+++ b/libs/cuabot/src/prompts/SYSTEM.md
@@ -1,175 +1,19 @@
You are a computer-use agent with your own Ubuntu 22.04 sandbox (xpra). User sees your desktop in real-time. You have sudo privileges.
## Environment
-- Node.js 20.x, Python 3.10, pip3, uv (Python package manager)
-- Currently installed: chromium, agent-browser, feh, claude-code, x11-apps
+- Node.js 22.x, Python 3.10, pip3, uv (Python package manager)
+- Currently installed: chromium, agent-browser, agent-device (adb), feh, claude-code, x11-apps
- Python libs: matplotlib, numpy, pandas, seaborn, plotly
- uv is available for running self-contained Python scripts with inline dependencies
+## Agent Tools
+- **computer-use**: Desktop automation MCP. Use for all screenshot tasks or desktop app automation.
+- **agent-browser**: Browser automation CLI. Use for all web browsing tasks.
+- **agent-device**: Android device/emulator automation CLI. Use for Android app testing and automation.
+
## Guidelines
- Keep responses brief
- Use agent-browser for browser automation (not chromium-browser)
-- Use bash tool to run agent-browser commands
+- Use agent-device for Android emulator/device automation
+- Use bash tool to run agent-browser and agent-device commands
- If asked to open an app that's not installed, install it first using `sudo apt-get install`
-
-## agent-browser
-
-Browser automation CLI for AI agents. Always use `agent-browser` when doing browser tasks, NEVER use `chromium-browser` or `firefox` unless specifically asked.
-
-### Quick Start
-
-```bash
-agent-browser open example.com
-agent-browser snapshot # Get accessibility tree with refs
-agent-browser click @e2 # Click by ref from snapshot
-agent-browser fill @e3 "[email protected]" # Fill by ref
-agent-browser get text @e1 # Get text by ref
-agent-browser screenshot page.png
-agent-browser close
-```
-
-### Core Commands
-
-```bash
-agent-browser open <url> --headed # Navigate to URL (use --headed to show browser)
-agent-browser click <sel> # Click element
-agent-browser dblclick <sel> # Double-click element
-agent-browser type <sel> <text> # Type into element
-agent-browser fill <sel> <text> # Clear and fill
-agent-browser press <key> # Press key (Enter, Tab, Control+a)
-agent-browser keydown <key> # Hold key down
-agent-browser keyup <key> # Release key
-agent-browser hover <sel> # Hover element
-agent-browser select <sel> <val> # Select dropdown option
-agent-browser check <sel> # Check checkbox
-agent-browser uncheck <sel> # Uncheck checkbox
-agent-browser scroll <dir> [px] # Scroll (up/down/left/right)
-agent-browser drag <src> <tgt> # Drag and drop
-agent-browser screenshot [path] # Take screenshot
-agent-browser snapshot # Accessibility tree with refs (best for AI)
-agent-browser close # Close browser
-```
-
-### Get Info
-
-```bash
-agent-browser get text <sel> # Get text content
-agent-browser get html <sel> # Get innerHTML
-agent-browser get value <sel> # Get input value
-agent-browser get attr <sel> <attr> # Get attribute
-agent-browser get title # Get page title
-agent-browser get url # Get current URL
-agent-browser get count <sel> # Count matching elements
-```
-
-### Check State
-
-```bash
-agent-browser is visible <sel> # Check if visible
-agent-browser is enabled <sel> # Check if enabled
-agent-browser is checked <sel> # Check if checked
-```
-
-### Find Elements (Semantic Locators)
-
-```bash
-agent-browser find role <role> <action> [value] # By ARIA role
-agent-browser find text <text> <action> # By text content
-agent-browser find label <label> <action> [value] # By label
-agent-browser find placeholder <ph> <action> [value] # By placeholder
-agent-browser find testid <id> <action> [value] # By data-testid
-```
-
-**Actions:** `click`, `fill`, `check`, `hover`, `text`
-
-**Examples:**
-```bash
-agent-browser find role button click --name "Submit"
-agent-browser find text "Sign In" click
-agent-browser find label "Email" fill "[email protected]"
-```
-
-### Wait
-
-```bash
-agent-browser wait <selector> # Wait for element to be visible
-agent-browser wait <ms> # Wait for time (milliseconds)
-agent-browser wait --text "Welcome" # Wait for text to appear
-agent-browser wait --url "**/dash" # Wait for URL pattern
-agent-browser wait --load networkidle # Wait for load state
-```
-
-### Mouse Control
-
-```bash
-agent-browser mouse move <x> <y> # Move mouse
-agent-browser mouse down [button] # Press button (left/right/middle)
-agent-browser mouse up [button] # Release button
-agent-browser mouse wheel <dy> [dx] # Scroll wheel
-```
-
-### Tabs & Navigation
-
-```bash
-agent-browser tab # List tabs
-agent-browser tab new [url] # New tab
-agent-browser tab <n> # Switch to tab n
-agent-browser tab close [n] # Close tab
-agent-browser back # Go back
-agent-browser forward # Go forward
-agent-browser reload # Reload page
-```
-
-### Snapshot Options
-
-```bash
-agent-browser snapshot # Full accessibility tree
-agent-browser snapshot -i # Interactive elements only
-agent-browser snapshot -c # Compact (remove empty elements)
-agent-browser snapshot -d 3 # Limit depth to 3 levels
-agent-browser snapshot -s "#main" # Scope to CSS selector
-agent-browser snapshot -i -c -d 5 # Combine options
-```
-
-### Selectors
-
-#### Refs (Recommended)
-
-Refs provide deterministic element selection from snapshots:
-
-```bash
-# 1. Get snapshot with refs
-agent-browser snapshot
-# Output:
-# - heading "Example Domain" [ref=e1] [level=1]
-# - button "Submit" [ref=e2]
-# - textbox "Email" [ref=e3]
-
-# 2. Use refs to interact
-agent-browser click @e2
-agent-browser fill @e3 "[email protected]"
-agent-browser get text @e1
-```
-
-#### CSS Selectors
-
-```bash
-agent-browser click "#id"
-agent-browser click ".class"
-agent-browser click "div > button"
-```
-
-#### Text & XPath
-
-```bash
-agent-browser click "text=Submit"
-agent-browser click "xpath=//button"
-```
-
-### Optimal AI Workflow
-
-1. Open page: `agent-browser open example.com`
-2. Get snapshot: `agent-browser snapshot -i`
-3. Identify target refs from snapshot
-4. Execute actions using refs: `agent-browser click @e2`
-5. Re-snapshot after page changes
diff --git a/libs/cuabot/src/scripts/agent-device-wrapper.sh b/libs/cuabot/src/scripts/agent-device-wrapper.sh
new file mode 100644
index 0000000000..0a3e0f9ab0
--- /dev/null
+++ b/libs/cuabot/src/scripts/agent-device-wrapper.sh
@@ -0,0 +1,179 @@
+#!/bin/bash
+# Wrapper for agent-device
+# System images are downloaded on-demand via "agent-device install"
+
+export ANDROID_SDK_ROOT=/home/user/android-sdk
+export JAVA_HOME=/usr/lib/jvm/java-17-openjdk-amd64
+export PATH="${ANDROID_SDK_ROOT}/cmdline-tools/latest/bin:${ANDROID_SDK_ROOT}/platform-tools:${ANDROID_SDK_ROOT}/emulator:${PATH}"
+
+# Check if nested virtualization is enabled
+check_kvm() {
+ if [ ! -e /dev/kvm ]; then
+ echo "=============================================="
+ echo "WARNING: Nested virtualization is not enabled!"
+ echo "=============================================="
+ echo ""
+ echo "agent-device requires KVM for Android emulation."
+ echo "Please restart cuabot with nested virtualization enabled:"
+ echo ""
+ echo " CUABOT_NESTED_VIRT=1 cuabot"
+ echo ""
+ echo "=============================================="
+ exit 1
+ fi
+}
+
+# Download and install the system image
+install_system_image() {
+ check_kvm
+
+ echo "[agent-device] Downloading Android system image..."
+ echo "[agent-device] This may take up to 15 minutes depending on your connection."
+ echo ""
+
+ # Accept licenses
+ yes | sdkmanager --licenses 2>/dev/null || true
+
+ # Download system image
+ echo "[agent-device] Downloading system-images;android-35;google_apis_playstore;x86_64..."
+ sdkmanager "system-images;android-35;google_apis_playstore;x86_64"
+
+ if [ $? -ne 0 ]; then
+ echo ""
+ echo "ERROR: Failed to download system image"
+ exit 1
+ fi
+
+ # Create AVD
+ echo ""
+ echo "[agent-device] Creating Android Virtual Device..."
+ echo "no" | avdmanager create avd -n device \
+ -k "system-images;android-35;google_apis_playstore;x86_64" -d "pixel_6" 2>/dev/null || true
+
+ # Configure AVD for container use
+ mkdir -p ~/.android/avd/device.avd
+ cat > ~/.android/avd/device.avd/config.ini << 'AVDEOF'
+avd.ini.encoding=UTF-8
+AvdId=device
+PlayStore.enabled=true
+abi.type=x86_64
+disk.dataPartition.size=4G
+hw.accelerator.isAccelerated=yes
+hw.cpu.arch=x86_64
+hw.cpu.ncore=4
+hw.gpu.enabled=yes
+hw.gpu.mode=swiftshader_indirect
+hw.keyboard=yes
+hw.lcd.density=420
+hw.lcd.height=2400
+hw.lcd.width=1080
+hw.ramSize=4096
+image.sysdir.1=system-images/android-35/google_apis_playstore/x86_64/
+tag.display=Google Play
+tag.id=google_apis_playstore
+vm.heapSize=576
+AVDEOF
+
+ echo ""
+ echo "[agent-device] Installation complete!"
+ echo "[agent-device] You can now use agent-device commands."
+ exit 0
+}
+
+# Handle install command
+if [ "$1" = "install" ]; then
+ install_system_image
+fi
+
+# Check if nested virtualization is enabled
+check_kvm
+
+# Check if Android system images exist
+if ! ls /home/user/android-sdk/system-images/android-* 1>/dev/null 2>&1; then
+ echo "=============================================="
+ echo "ERROR: Android system images not found!"
+ echo "=============================================="
+ echo ""
+ echo "Run the following command to download the system image:"
+ echo ""
+ echo " agent-device install"
+ echo ""
+ echo "(This may take up to 15 minutes)"
+ echo "=============================================="
+ exit 1
+fi
+
+# Create AVD if it doesn't exist
+if [ ! -d ~/.android/avd/device.avd ]; then
+ echo "[agent-device] Creating Android Virtual Device..."
+
+ # Find the installed system image
+ SYSTEM_IMAGE=$(ls -d ${ANDROID_SDK_ROOT}/system-images/android-*/google_apis_playstore/x86_64 2>/dev/null | head -1)
+ if [ -z "$SYSTEM_IMAGE" ]; then
+ SYSTEM_IMAGE=$(ls -d ${ANDROID_SDK_ROOT}/system-images/android-*/google_apis/x86_64 2>/dev/null | head -1)
+ fi
+ if [ -z "$SYSTEM_IMAGE" ]; then
+ SYSTEM_IMAGE=$(ls -d ${ANDROID_SDK_ROOT}/system-images/android-*/default/x86_64 2>/dev/null | head -1)
+ fi
+
+ if [ -z "$SYSTEM_IMAGE" ]; then
+ echo "ERROR: Could not find a valid system image in ${ANDROID_SDK_ROOT}/system-images/"
+ exit 1
+ fi
+
+ # Extract the image identifier
+ ANDROID_VER=$(echo "$SYSTEM_IMAGE" | grep -oP 'android-\d+' | head -1)
+ IMAGE_TYPE=$(basename $(dirname "$SYSTEM_IMAGE"))
+ IMAGE_ID="system-images;${ANDROID_VER};${IMAGE_TYPE};x86_64"
+
+ echo "[agent-device] Using system image: $IMAGE_ID"
+ echo "no" | avdmanager create avd -n device -k "$IMAGE_ID" -d "pixel_6" 2>/dev/null || true
+
+ # Configure AVD for container use
+ mkdir -p ~/.android/avd/device.avd
+ IMAGE_SYSDIR=$(echo "$SYSTEM_IMAGE" | sed "s|${ANDROID_SDK_ROOT}/||")
+
+ cat > ~/.android/avd/device.avd/config.ini << AVDEOF
+avd.ini.encoding=UTF-8
+AvdId=device
+PlayStore.enabled=true
+abi.type=x86_64
+disk.dataPartition.size=4G
+hw.accelerator.isAccelerated=yes
+hw.cpu.arch=x86_64
+hw.cpu.ncore=4
+hw.gpu.enabled=yes
+hw.gpu.mode=swiftshader_indirect
+hw.keyboard=yes
+hw.lcd.density=420
+hw.lcd.height=2400
+hw.lcd.width=1080
+hw.ramSize=4096
+image.sysdir.1=${IMAGE_SYSDIR}/
+tag.display=Google Play
+tag.id=${IMAGE_TYPE}
+vm.heapSize=576
+AVDEOF
+
+ echo "[agent-device] AVD created!"
+fi
+
+# Fix KVM permissions if needed
+if [ -e /dev/kvm ] && [ ! -w /dev/kvm ]; then
+ sudo chmod 666 /dev/kvm 2>/dev/null || true
+fi
+
+# Auto-start emulator if no Android devices are connected
+if ! ${ANDROID_SDK_ROOT}/platform-tools/adb devices 2>/dev/null | grep -q "emulator\|device$"; then
+ echo "[agent-device] Starting Android emulator..."
+ nohup ${ANDROID_SDK_ROOT}/emulator/emulator -avd device -no-audio -gpu swiftshader_indirect -no-snapshot -no-boot-anim &>/tmp/emulator.log &
+ ${ANDROID_SDK_ROOT}/platform-tools/adb wait-for-device
+ echo "[agent-device] Waiting for boot..."
+ while [ "$(${ANDROID_SDK_ROOT}/platform-tools/adb shell getprop sys.boot_completed 2>/dev/null)" != "1" ]; do
+ sleep 2
+ done
+ echo "[agent-device] Emulator ready!"
+fi
+
+# Run the real agent-device binary
+exec /usr/lib/node_modules/agent-device/bin/agent-device.mjs "$@"
diff --git a/libs/cuabot/src/scripts/entrypoint.sh b/libs/cuabot/src/scripts/entrypoint.sh
new file mode 100644
index 0000000000..ba48b496f9
--- /dev/null
+++ b/libs/cuabot/src/scripts/entrypoint.sh
@@ -0,0 +1,18 @@
+#!/bin/bash
+# Cuabot container entrypoint
+# Adds skills on startup (they go to /home/user/.claude which is mounted)
+
+# Add skills in background if not already present
+if [ ! -f /home/user/.claude/.skills-installed ]; then
+ (
+ echo "[entrypoint] Installing skills in background..."
+ npx skills add callstackincubator/agent-device --all 2>/dev/null || true
+ npx skills add vercel-labs/agent-browser --all 2>/dev/null || true
+ touch /home/user/.claude/.skills-installed
+ echo "[entrypoint] Skills installed"
+ ) &
+fi
+
+# Start dbus and xpra
+dbus-daemon --session --fork --address=unix:path=/tmp/dbus-session 2>/dev/null || true
+exec xpra seamless :100 --sharing=yes --no-daemon --bind-tcp=0.0.0.0:10000 --html=on --dpi=96