The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/gemini_tts.py utilizes subprocess.run to call the gcloud CLI tool to automatically detect and retrieve the project ID for Google Cloud/Vertex AI authentication.
[DATA_EXFILTRATION]: The voice generation scripts (elevenlabs.py, gemini_tts.py, and openai_tts.py) access sensitive file paths including ~/.env and ~/.config/skills/.env to load API credentials for authentication. This access is limited to retrieving keys for the legitimate operation of the text-to-speech services.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to the lack of input sanitization in the voice synthesis process.
Ingestion points: Untrusted data enters the skill via the --text and --style arguments in scripts/gemini_tts.py.
Boundary markers: Content and instructions are concatenated without the use of protective delimiters or 'ignore' warnings.
Capability inventory: The skill utilizes the Google Gemini API to generate audio content based on provided prompts.
Sanitization: Input text is processed directly without escaping or validation before being passed to the multimodal generation endpoint.

voice-generation