The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py executes external binaries (apktool and jadx) via subprocess.run. The arguments are passed safely as lists, and shell=True is omitted, mitigating standard shell injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection via untrusted third-party inputs:
Ingestion points: scripts/agent.py walks through decompiled directory structures and reads code from unverified third-party Java files generated from the analyzed APK.
Boundary markers: The script prints out regex-matched text snippets directly without wrapping them in explicit text delimiters or instructing the agent to ignore any embedded directives.
Capability inventory: The skill possesses capabilities to read files and execute binaries via subprocess.run inside scripts/agent.py.
Sanitization: Content from the malware files is truncated via string slicing ([:80]) but is not escaped or sanitized against prompt injection patterns.

reverse-engineering-android-malware-with-jadx