The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to download Android APK files from well-known third-party repositories, including APKPure and APKMirror. It also fetches the jadx decompilation tool directly from its official GitHub repository. These sources are recognized as well-known or trusted within the security research community for the intended purpose.
[COMMAND_EXECUTION]: The instructions involve executing various command-line utilities for archive extraction (unzip, 7z), decompilation (jadx, apktool), and runtime instrumentation (frida, objection, adb). These tools are standard for the mobile application auditing workflow described.
[DATA_EXFILTRATION]: Extensive scanning patterns are provided to identify sensitive data, such as API keys (AWS, Google, GitHub, Slack, OpenAI, etc.) and JWT tokens, within target applications. This is part of the intended target analysis; no evidence was found of exfiltrating the agent's host environment data or hardcoding sensitive credentials belonging to the skill itself.
[PROMPT_INJECTION]: The skill processes external, untrusted data (APK files) using tools like jadx and strings.
Ingestion points: Untrusted Android APK files are downloaded via curl and processed locally (SKILL.md).
Boundary markers: Absent; the skill does not explicitly isolate the analysis environment from potentially malicious payloads within the APKs.
Capability inventory: The skill uses a broad set of tools including curl, wget, unzip, 7z, jadx, apktool, frida, objection, and adb, which could be targeted by a malicious binary.
Sanitization: No specific sanitization or sandboxing of the APK content is mentioned before analysis.
This vulnerability surface is inherent to the skill's primary purpose of security auditing.

apk-redteam-pipeline