The Agent Skills Directory

[COMMAND_EXECUTION]: The skill dynamically identifies and executes a local script called tzai-image located in various home directory subfolders (e.g., .agents, .cursor). It uses the bash shell to run this engine with command-line arguments.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to how it handles user input.
Ingestion points: The <user subject> placeholder in SKILL.md is populated with raw user input.
Boundary markers: The input is wrapped in double quotes in the bash execution block, but there are no instructions to the agent to escape shell metacharacters.
Capability inventory: The skill utilizes bash execution which can be exploited if the interpolated input contains command separators or subshells.
Sanitization: No sanitization, filtering, or validation steps are provided for the user-supplied content before it is passed to the shell.

tzai-onboarding