The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains several PowerShell and Bash scripts (e.g., decode.ps1, frida-run.ps1, rebuild-sign-install.ps1) that execute local CLI tools such as jadx, apktool, frida, and adb. This behavior is the primary intended function of the skill for performing APK analysis.
[EXTERNAL_DOWNLOADS]: The SKILL.md file describes an 'On-Demand Bootstrap' system that automatically fetches missing tools. Scripts like decode.ps1 and frida-run.ps1 call a bootstrap script to download jadx and apktool from GitHub releases and install frida-tools via PyPI. These sources are well-known technology platforms.
[CREDENTIALS_UNSAFE]: The script rebuild-sign-install.ps1 uses a hardcoded debug keystore (debug.keystore) with the default password 'android'. This is standard industry practice for development and reverse engineering environments and does not constitute a leak of sensitive user credentials.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through the analysis of APK binary content, AndroidManifest.xml files, and smali source code.
Boundary markers: The analysis scripts do not implement explicit boundary markers or instructions to the agent to ignore embedded commands within the processed files.
Capability inventory: The skill has significant capabilities including subprocess execution of reverse engineering tools, file system writes (output directories), and network/device operations via adb and frida.
Sanitization: There is minimal sanitization of data extracted from the APKs, such as basic filename normalization in decode.ps1.

apk-reverse