The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs users to download binaries and source code from individual GitHub repositories that are not verified organizations.
Evidence: https://github.com/CodingGay/BlackDex/releases for downloading the BlackDex APK.
Evidence: https://github.com/Youlor/Youpk for downloading the Youpk ROM and associated tools.
[REMOTE_CODE_EXECUTION]: Instructs the execution of downloaded third-party binaries and packages.
Evidence: java -jar dexfixer.jar ./unpacker ./out runs a Java archive obtained from an untrusted GitHub source.
Evidence: pip install frida-dexdump installs an external package that is not part of the core trusted Frida organization.
[COMMAND_EXECUTION]: Executes commands that modify the Android environment and interact with running applications.
Evidence: adb shell "echo <包名> >> /data/local/tmp/unpacker.config" writes configuration data to a system-wide temporary directory.
Evidence: adb shell monkey -p <包名> 1 is used to programmatically launch target applications.
[DYNAMIC_EXECUTION]: Employs dynamic instrumentation to modify and observe application behavior at runtime.
Evidence: Use of frida and frida-dexdump to search, attach to, and extract DEX files from active process memory.
Evidence: Mentions of hooking system APIs and DexFile cookie techniques for extraction.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Android application data, creating an attack surface for instructions embedded within the target files.
Ingestion points: Target APK and DEX files processed by jadx and dexdump as specified in SKILL.md.
Boundary markers: None present; the agent processes raw output from decompilation and dumping tools directly.
Capability inventory: Includes adb shell access, pip package installation, and execution of java binaries.
Sanitization: Absent; the data retrieved from the target applications is used to drive the next steps of the analysis without verification.

re-mobile-pack