The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a shell script to dynamically locate an executable engine by iterating through several hidden directories in the user's home path, including ~/.agents/, ~/.claude/, ~/.cursor/, and others. The identified script is then executed using bash "$ENGINE".
[PROMPT_INJECTION]: User-provided text is directly interpolated into a shell command through the --prompt "<user subject>" argument. This represents an indirect prompt injection surface where malicious user input could potentially attempt to break out of the command string if not properly sanitized by the underlying engine.
[EXTERNAL_DOWNLOADS]: The skill references and provides instructions for installing an external dependency from the author's repository using the command npx skills add kedoupi/tzai-image-skill. It also references a gallery hosted on GitHub at https://github.com/kedoupi/tzai-image-skill.
[INDIRECT_PROMPT_INJECTION]: The skill fulfills the requirements for indirect prompt injection vulnerability:
Ingestion points: User input provided via the /tzai-infographic slash command (SKILL.md).
Boundary markers: None identified; input is wrapped in double quotes in a bash script but not otherwise sanitized.
Capability inventory: Execution of shell scripts and passing of arguments to a CLI tool (SKILL.md).
Sanitization: None detected in the provided skill file.

tzai-infographic