The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill automates the setup of security testing environments by downloading necessary binaries and tools from trusted official sources. Specifically:
Fetches Frida server releases from its official GitHub repository in setup/frida_server_manager.py and setup/magisk_module/service.sh.
Downloads the Magisk rooting utility from the developer's official GitHub repository in setup/setup_emulator.py.
Fetches the Drozer security framework from WithSecureLabs' GitHub repository in setup/setup_physical_device.sh.
[COMMAND_EXECUTION]: The skill relies heavily on executing system commands through subprocess.run and subprocess.Popen to interact with the Android Debug Bridge (ADB) and the Frida instrumentation engine. This behavior is fundamental to the skill's primary function of auditing mobile applications and extracting diagnostic data from connected devices.
[DYNAMIC_EXECUTION]: The orchestration logic in workflows/pentest_runner.py uses importlib to dynamically load security scanning modules based on user requests. Additionally, the skill frequently generates and injects JavaScript hooks into running Android processes at runtime to monitor cryptographic and authentication functions.
[INDIRECT_PROMPT_INJECTION]: The skill has a notable surface for indirect prompt injection due to its automated data collection features:
Ingestion points: Workflow scripts like workflows/data_leakage_scanner.py and workflows/full_assessment.py programmatically ingest potentially untrusted data such as application logs (logcat), local database contents, and SharedPreferences files.
Boundary markers: The scripts often lack explicit delimiters when processing this external data, though they generally process it for structured reporting rather than direct prompt re-injection.
Capability inventory: The skill possesses high-privilege capabilities including arbitrary ADB shell command execution, file system access on the host, and network communication via proxy.
Sanitization: While many system calls use list-based arguments to prevent command injection on the host, some fallback logic uses shell execution on external strings.

android-pentest