The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the ffmpeg command-line utility via subprocess.run to perform audio speed normalization using the atempo filter. Evidence is found in the change_audio_speed function within SKILL.md. The execution is restricted to temporary file paths and numeric factors, which reduces risk.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text to generate audio and synchronization metadata, creating an ingestion surface for untrusted data. It implements regex-based cleaning (re.sub(r'<[^>]*>', '', text)) and explicit string replacements to sanitize input before processing.\n- [SAFE]: The skill uses torch.load with weights_only=True to load speaker embeddings, which is a recommended security practice to prevent code execution via malicious serialized data.

chattts-voice-generation