The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to interact with the system via shell commands using the agent-voice CLI. This includes tool setup (agent-voice setup) and speech generation (agent-voice speak). While these are standard for CLI integration, they represent the primary execution path for the skill.
[EXTERNAL_DOWNLOADS]: The skill guides the agent to install the agent-voice package using uv tool install. This involves downloading code from a remote repository. The source is identified as the author's own GitHub repository (yoav0gal/agent-voice), representing a legitimate vendor resource.
[INDIRECT_PROMPT_INJECTION]: The skill processes text from the conversation context and interpolates it into shell commands, which presents a potential vulnerability surface if the content is not sanitized.
Ingestion points: The RESPONSE_AS_TEXT variable, which is populated with content generated or retrieved during the AI conversation, is used as an argument for the agent-voice tool.
Boundary markers: The skill suggests using double quotes ("$RESPONSE_AS_TEXT") for shell arguments, which provides basic shell word-splitting protection but does not prevent all forms of command injection (e.g., via backticks or command substitution sequences within the variable).
Capability inventory: The skill utilizes shell execution, file system writes for temporary audio files, and rendering of HTML/Markdown content.
Sanitization: There are no explicit instructions for the agent to sanitize or escape shell metacharacters (e.g., ;, |, &, $, `) before execution.

create-speech-recording-desktop