The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided text for speech generation, creating a surface for indirect prompt injection if the source text is untrusted. 1. Ingestion points: Text content provided via --text or --file arguments, and style directions via --style in gemini_tts.py, elevenlabs.py, and openai_tts.py. 2. Boundary markers: Absent; user-provided text is passed directly to external APIs. 3. Capability inventory: Network access to third-party AI APIs (Google, OpenAI, ElevenLabs) and file system write access to save generated audio files. 4. Sanitization: None; the skill performs no filtering on the input text.\n- [COMMAND_EXECUTION]: The scripts/gemini_tts.py script executes a shell command to auto-detect the Google Cloud environment. Evidence: The script uses subprocess.run(['gcloud', 'config', 'get-value', 'project']) with a static argument list to identify the active project for Vertex AI authentication. This is a benign discovery mechanism for environment configuration.

voice-generation