The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts scripts/agent.py and scripts/process.py execute the external command-line utility aapt using subprocess.run() to inspect Android package permissions. The command is structured safely as a list of individual arguments without enabling the system shell, minimizing shell injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an inherent attack surface for indirect prompt injection due to processing external, untrusted application components and metadata.
Ingestion points: scripts/agent.py and scripts/process.py extract code patterns, raw permissions, string fields from uploaded mobile files, and receive analysis results from the external VirusTotal API.
Boundary markers: There are no structural boundary markers or instructions configured to insulate the agent prompt environment from untrusted text embedded within application contents or threat intelligence feeds.
Capability inventory: The skill is equipped with shell command execution capabilities via subprocess.run and network connectivity via the requests library.
Sanitization: Extracted code strings, dex data samples, and external vendor report names are not sanitized or escaped before presentation to the agent context.

detecting-mobile-malware-behavior