adb-claw

Fail

Audited by Gen Agent Trust Hub on Jul 17, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill's frontmatter configuration initiates the download of pre-compiled executable binaries (adb-claw) for Darwin and Linux architectures from a third-party GitHub repository (llm-net/adb-claw).
  • [REMOTE_CODE_EXECUTION]: The skill is designed to execute the downloaded adb-claw binary on the host machine to manage the connected Android device. Additionally, the monitor and audio capture commands push and execute temporary DEX helper files onto the Android device at runtime.
  • [COMMAND_EXECUTION]: The skill provides a shell command that enables the agent to run arbitrary shell commands on the connected Android device. This high-privilege capability allows for full device manipulation, including changing system settings and accessing private storage, if the agent's instructions are compromised.
  • [DATA_EXFILTRATION]: The skill contains features for broad data access on the mobile device, including screenshot, audio capture for system audio, and monitor for live UI text. These capabilities allow the agent to capture and potentially exfiltrate sensitive user information appearing on the screen or heard through the speakers.
  • [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection through the processing of untrusted device data.
  • Ingestion points: UI text and chat messages are ingested via the monitor, observe, and live cart commands in SKILL.md.
  • Boundary markers: Absent. The instructions do not specify delimiters or warnings for the agent to ignore instructions embedded in the device's UI text.
  • Capability inventory: The skill provides access to shell, file transfer, app management, and UI interaction (tap, swipe) commands in SKILL.md.
  • Sanitization: Absent. The data retrieved from the accessibility framework is passed to the agent without filtering or escaping.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jul 17, 2026, 10:01 AM
Security Audit — agent-trust-hub — adb-claw