The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/preflight-validation.py uses subprocess.run with shell=True to execute ADB commands. The device identifier provided via the --device CLI flag is interpolated directly into the shell command string (e.g., full_cmd = f"adb -s {device_id} shell ...") without sanitization or escaping. This allows for arbitrary command execution on the host machine if the device_id parameter is maliciously crafted (e.g., 127.0.0.1:5555; touch /tmp/pwned).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from an external source (a connected Android device) via adb shell commands such as getprop, pm list packages, and df.
Ingestion points: scripts/preflight-validation.py reads device properties and package lists; bypass-validation.toon and integrity-verification.toon perform similar checks via the adb-shell action.
Boundary markers: None. The outputs from these commands are processed directly as strings.
Capability inventory: The skill has the capability to execute shell commands via subprocess.run and write to the local filesystem (e.g., adb-screen-capture in bypass-validation.toon).
Sanitization: There is no significant sanitization of the data retrieved from the device before it is processed or printed. A malicious device could potentially return crafted property values or package names intended to influence the agent's logic or output.

adb-bypass