The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user/agent to download and execute scripts directly from a remote community repository using the frida --codeshare command (specifically referencing akabe1/frida-multiple-unpinning). This command fetches and executes unverified third-party code at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data from mobile application binaries and their contents.
Ingestion points: The agent is instructed to analyze AndroidManifest.xml, strings.xml, and decompiled smali/Java source code using grep and jadx.
Boundary markers: There are no instructions to use delimiters or ignore potential malicious instructions embedded within the metadata or strings of the analyzed APK/IPA files.
Capability inventory: The skill leverages powerful system capabilities including adb for device control, apktool for binary modification, and pip for package installation.
Sanitization: The skill lacks instructions for sanitizing or escaping content extracted from external binaries before processing it in the agent context.
[COMMAND_EXECUTION]: The skill involves the execution of multiple command-line utilities for setup and analysis, such as adb, apktool, jadx, and pip. While these are standard tools for the stated purpose, they represent a significant capability set if misused by an attacker controlling the input files.

mobile-pentest