The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local text-to-speech utility located at /opt/hatch/bin/tts using the muse.exec tool.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing user-supplied text to generate audio. It implements sanitization by using the --text-stdin flag to avoid shell-level injection and provides instructional guidance for formatting input as "spoken form" to reduce the risk of the agent interpreting embedded syntax. The skill also explicitly instructs the agent to treat data from voice configuration files (user/voices.json) as identifiers rather than instructions, mitigating schema confusion risks.
[PERSISTENCE]: The skill includes instructions for scheduling task retries via cron or delayed wakeups in the event of backend failures. These retries are bounded to a specific one-hour window and are restricted to the primary task purpose of audio synthesis, rather than attempts to maintain unauthorized persistent access.

tts