The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing untrusted data from the user that is then interpolated into shell commands.
Ingestion points: User-provided package names ($PKG) and JSON payloads for data safety and recovery actions defined in SKILL.md.
Boundary markers: Absent. There are no explicit instructions for the agent to validate or escape these inputs before execution.
Capability inventory: The skill executes the gps CLI tool which can modify app safety labels and recovery configurations.
Sanitization: Absent. The variables are interpolated directly into shell command strings without escaping logic.
[COMMAND_EXECUTION]: The skill constructs and executes shell commands using string interpolation of variables like $PKG.
Evidence: The command gps --package $PKG --confirm call gps_set_data_safety '{"packageName":"'"$PKG"'", ...}' in SKILL.md uses shell concatenation that is vulnerable if the input contains special characters.

gps-compliance