The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill encourages the installation of additional functionality from external sources.
It directs the agent to install the android-reverse-engineering plugin from an individual third-party GitHub repository (https://github.com/SimoneAvogadro/android-reverse-engineering-skill) using a skill-management CLI.
It also references official Android Open Source Project (AOSP) repositories at https://github.com/android/skills for performance analysis tools.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface where the agent processes untrusted data which could contain malicious instructions.
Ingestion points: The agent is instructed to read and analyze adb logcat output, ANR traces, heap dumps, Perfetto traces, and layout JSON dumps generated by the android layout tool. These sources can contain strings or metadata from third-party applications.
Boundary markers: There are no instructions to use delimiters or ignore potential embedded commands within the log data or traces.
Capability inventory: The skill utilizes powerful tools including adb for device interaction, ./gradlew for executing build scripts, and java for running JAR files.
Sanitization: No sanitization or validation of the ingested log and trace content is performed before the agent processes it.
[COMMAND_EXECUTION]: The skill performs several direct command-line operations that interact with the host system and connected Android devices.
Frequent use of adb to pull sensitive system files such as traces and heap dumps from /data/anr/ and /data/local/tmp/.
Execution of Gradle tasks via ./gradlew, which runs arbitrary code defined in the project's build configuration.
Usage of java -jar to execute the R8/ProGuard retrace utility on local files.

android-debugging