The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes the uvx command, which downloads and executes the zai-tts package from an external registry (likely PyPI) at runtime. This introduces a dependency on an unverified third-party package.
[COMMAND_EXECUTION]: The skill performs shell command execution using uvx zai-tts. It constructs commands that include user-supplied data, which can lead to unintended execution if the input is not handled correctly.
[INDIRECT_PROMPT_INJECTION]: The skill instructions provide a template for shell commands where user-supplied messages are interpolated: uvx zai-tts -t "{msg}". This creates a command injection surface if the {msg} content is sourced from untrusted data (e.g., a summarized web page).
Ingestion points: The {msg} parameter in the usage examples in SKILL.md.
Boundary markers: The {msg} variable is enclosed in double quotes, but there is no instruction for the agent to escape shell-metacharacters (e.g., ;, &, |, $(...)) within that input.
Capability inventory: The skill has the capability to execute shell commands and write files to a temporary directory.
Sanitization: No sanitization or validation logic is defined to prevent an attacker from escaping the quotes and executing arbitrary shell commands.

zai-tts