The Agent Skills Directory

[COMMAND_EXECUTION]: The skill operates by executing the listenhub CLI tool to generate speech from text. It uses shell commands to interact with the binary, passing parameters for voice selection, language, and speed. It also utilizes standard system utilities like jq for JSON parsing and mkdir/cat for managing configuration files.
[EXTERNAL_DOWNLOADS]: After speech generation, the skill identifies an audioUrl from the service response and uses curl to download the MP3 file to the user's current directory. These downloads are directed to the vendor's infrastructure as part of the core feature set.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect injection by ingesting untrusted user text and interpolating it into shell commands or writing it to temporary files (/tmp/lh-content.txt).
Ingestion points: User-provided text strings in SKILL.md (Step 1 of Quick Mode and Step 5 of Script Mode).
Boundary markers: The skill suggests using quoted heredocs (<< 'ENDCONTENT') for long text, which provides a shell-level boundary, but no natural language boundary markers are explicitly instructed for the LLM prompt itself.
Capability inventory: File system writes (current directory and /tmp), network requests via curl, and execution of the listenhub binary.
Sanitization: No explicit sanitization or filtering of the input text is performed before it is passed to the generation tool.
[DYNAMIC_EXECUTION]: The skill dynamically constructs the command prefix ($CMD_PREFIX) based on the authentication state and generates variable output filenames based on the input topic to prevent overwriting existing files.

tts