The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands for text-to-speech generation and playback, incorporating variables such as $RESPONSE_AS_TEXT, $LABEL, $PATH, and $path directly into command lines.
Evidence: agent-voice speak "$RESPONSE_AS_TEXT" in SKILL.md and agent-voice play "$path" in references/delivery/default.md.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied text and interpolates it into shell commands. If the input contains shell metacharacters and the agent does not perform strict escaping, it could lead to arbitrary command execution.
Ingestion points: User-provided text or requested summaries used to populate the RESPONSE_AS_TEXT variable in SKILL.md.
Boundary markers: None present to delimit untrusted data from the command context or warn the agent about embedded shell syntax.
Capability inventory: The skill invokes the agent-voice CLI and uses uv for package management.
Sanitization: No instructions are provided for the agent to sanitize, escape, or validate the text before passing it to the shell.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the agent-voice tool from an external source.
Evidence: uv tool install agent-voice in SKILL.md.
Note: The tool originates from the skill author's repository ('yoav0gal') and is considered a vendor-owned resource.

create-speech-recording