The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill includes an automated "bootstrap" mechanism in its core scripts (e.g., scripts/decode.ps1 and scripts/frida-run.ps1) that executes external files like bootstrap-reverse.ps1 using ExecutionPolicy Bypass to facilitate tool installation if they are missing.
[EXTERNAL_DOWNLOADS]: The skill is configured to automatically fetch and install dependencies such as jadx, apktool, and frida-tools (via pip) from their respective official GitHub repositories and public registries during the self-bootstrapping phase.
[COMMAND_EXECUTION]: The skill facilitates the execution of powerful system-level CLI tools including adb for device control, frida for dynamic process instrumentation, and apktool/jadx for filesystem modification and decompilation. This is consistent with its intended purpose as a reverse engineering extension.
[PROMPT_INJECTION]: The skill has an indirect prompt injection attack surface (Category 8) because it instructs the agent to read and interpret decompiled source code from potentially untrusted APK files. Maliciously crafted APKs could embed instructions in Smali comments, Java string constants, or manifest fields to trick the AI agent into performing unauthorized actions.
Ingestion points: Decompiled source files and XML manifests processed by jadx and apktool in the jadx/ and apktool/ task directories.
Boundary markers: Absent; the agent is directed to read the decompiled output directly without specific delimiters or warnings to disregard embedded natural language instructions.
Capability inventory: The agent has capabilities for broad filesystem access, process injection via frida, and device-level operations via adb (shell, install, push/pull).
Sanitization: No sanitization is performed on the decompiled code before it is presented to the agent for logical analysis.

apk-reverse