The Agent Skills Directory

[DATA_EXPOSURE]: The skill utilizes adb shell dumpsys and adb bugreport to gather device information. These system-level diagnostics can contain sensitive data such as account identifiers, notification content, network configurations, and application logs. The skill provides clear instructions to redact this information and requires user authorization before generating broad diagnostic archives.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external Android devices via ADB. This output constitutes untrusted data that could potentially contain malicious instructions aimed at the AI agent. The following evidence chain applies:
Ingestion points: Data enters the agent context through the output of various adb shell dumpsys and top commands (documented in SKILL.md and references/evidence-and-interpretation.md).
Boundary markers: The skill explicitly instructs the agent to distinguish evidence from inference and implements a "Diagnostic Contract" and "Gate every intervention" policy.
Capability inventory: The agent has the capability to execute shell commands via ADB, though it is instructed to prioritize read-only commands.
Sanitization: The skill mandates the redaction of serial numbers, phone numbers, SSIDs, and account identifiers before sharing logs.
[COMMAND_EXECUTION]: The skill generates shell commands for execution on a connected Android device through adb. These are limited to standard diagnostic tools. The instructions explicitly prohibit the use of root privileges or the installation of external binaries like BusyBox, adhering to the principle of least privilege for the diagnostic task.

diagnose-android-overheating