The Agent Skills Directory

[PROMPT_INJECTION]: The script scripts/generate.py is susceptible to indirect prompt injection as it interpolates external text directly into the AI prompt template without protective delimiters.
Ingestion points: Text is ingested from command-line arguments or local files via the -f/--file argument in scripts/generate.py.
Boundary markers: Absent. The input is directly concatenated into a string template (Say {style}: {text}) before being sent to the model.
Capability inventory: The script performs network requests to the Gemini API and writes audio files to the local file system.
Sanitization: None. The script does not filter or escape the input text for potential instructions directed at the TTS model.
[COMMAND_EXECUTION]: The script scripts/generate.py provides a mechanism to disable SSL certificate verification, which is a significant security weakness.
Evidence: The --no-ssl-verify flag and the GEMINI_TTS_NO_SSL_VERIFY environment variable trigger the disable_ssl_verification() function.
Method: This function modifies the global ssl state and monkey-patches the httpx library to skip certificate validation, increasing the risk of man-in-the-middle (MitM) attacks during API communication.

gemini-tts