adb controller

• Outcome: Review
• Latest audit: Sep 11, 2026 · 4:57 AM UTC
• Version: 1.0.0

Overview

This Android ADB controller is mostly coherent, but it needs Review because it permits broad device control, silently relies on an environment-selected target, and saves screenshots after each command.

Install only if you are comfortable giving the agent broad ADB control over the target Android device. Before use, verify which device ADB will target, avoid setting ADB_SERVER_ADDRESS to untrusted hosts, and assume screenshots saved in ~/.openclaw/workspace may contain sensitive screen contents until deleted.

A.I.G

Vulnerability Patterns

• Skill Instruction Hijacking: Alters the agent's session goals or safety constraints when the skill loads
• Agent Memory Poisoning: Writes attacker-controlled rules into memory that affect later sessions
• Remote Payload Retrieval and Execution: Fetches external code whose behavior can change after review
• Embedded Malicious Code: Ships malicious scripts inside the skill and executes them locally
• Unauthorized Access and Privilege Escalation: Obtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Behavioral AST: exec() Call, eval() Call, Dynamic Import
• Taint Tracking: Direct Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration

Findings (9)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 93% confidence
• Finding: The skill explicitly instructs the agent to execute a local Python script via the exec tool and states that it reads configuration from openclaw.json, yet the manifest declares no tool scope or allowed-tools restrictions. In a skill that can issue arbitrary ADB arguments to a connected device, missing tool restrictions weakens containment and increases the chance of unintended shell access, unsafe invocation, or broader-than-expected execution capability.

Vague Triggers

Medium

• Confidence: 96% confidence
• Finding: The description says to use this skill whenever the user asks to control an Android device, tap, swipe, input text, or perform actions via adb, which is a broad trigger for a highly capable action skill. Because the skill can pass arbitrary ADB arguments and affect a real device, broad activation increases the risk of accidental invocation, misuse for sensitive device operations, or execution without sufficient user confirmation for destructive actions.

Description-Behavior Mismatch

Medium

• Confidence: 96% confidence
• Finding: The implementation diverges from the manifest by taking the ADB target from an environment variable instead of the declared openclaw.json configuration, weakening operator expectations and configuration integrity. This mismatch can make the skill easier to retarget silently, increasing the chance of unauthorized device control or data capture.

Missing User Warnings

Medium

• Confidence: 89% confidence
• Finding: Reading ADB_SERVER_ADDRESS from the environment and auto-connecting to a host:port can create an undisclosed outbound network connection to a remote Android device or emulator. In this skill context, that is more dangerous because it directly precedes privileged device-control operations, so a hidden retargeting of the ADB endpoint can expose both commands and collected screenshots.

subprocess module call

Medium

• Category: Dangerous Code Execution

Content

if adb_server:
        # If it looks like an IP address/port, attempt to connect first
        if ":" in adb_server:
            subprocess.run(["adb", "connect", adb_server], capture_output=True)
        base_cmd.extend(["-s", adb_server])
        
    cmd = base_cmd + args

• Confidence: 90% confidence
• Finding: The script passes an environment-derived adb_server value directly into an adb connect operation, allowing external configuration to trigger network connections to attacker-controlled hosts or unintended devices. Although this is not shell injection, it can redirect the skill to a rogue ADB endpoint and cause commands and screenshots to be sent to the wrong target.

Tainted flow: 'adb_server' from os.environ.get (line 15, credential/environment) → subprocess.run (code execution)

Medium

• Category: Data Flow

Content

if adb_server:
        # If it looks like an IP address/port, attempt to connect first
        if ":" in adb_server:
            subprocess.run(["adb", "connect", adb_server], capture_output=True)
        base_cmd.extend(["-s", adb_server])
        
    cmd = base_cmd + args

• Confidence: 94% confidence
• Finding: Untrusted data from ADB_SERVER_ADDRESS flows into a code path that initiates an ADB connection, enabling environment manipulation to change the device target or establish outbound connections. In this skill, that is especially sensitive because subsequent device-control commands and screenshot capture may operate on an unintended or malicious endpoint.

subprocess module call

Medium

• Category: Dangerous Code Execution

Content

print(f"Running: {' '.join(cmd)}")
    
    try:
        result = subprocess.run(cmd, check=True, capture_output=True, text=True)
        print(result.stdout)
        if result.stderr:
            print(result.stderr, file=sys.stderr)

• Confidence: 70% confidence
• Finding: subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Missing User Warnings

Medium

• Confidence: 96% confidence
• Finding: The script automatically captures and stores a device screenshot after every ADB command, which can persist sensitive screen contents such as messages, credentials, health data, or tokens to disk without user awareness. In an Android-control skill, this materially increases privacy risk because screenshots are taken routinely and retained in a predictable workspace directory.

subprocess module call

Medium

• Category: Dangerous Code Execution

Content

screencap_cmd = base_cmd + ["exec-out", "screencap", "-p"]
    try:
        with open(screenshot_path, "wb") as f:
            subprocess.run(screencap_cmd, stdout=f, check=True)
        print(f"Screenshot saved to: {screenshot_path}")
    except Exception as e:
        print(f"Failed to take screenshot: {e}", file=sys.stderr)

• Confidence: 70% confidence
• Finding: subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Static analysis

No suspicious patterns detected.