The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes transcription text generated from external audio and video files, which constitutes untrusted third-party data. This creates a surface for Indirect Prompt Injection where malicious instructions could be embedded in the audio recordings. The skill proactively mitigates this risk by providing clear instructions to the agent to treat transcript content exclusively as data, wrap excerpts in specific quote blocks, and ignore any directives found within the text.
[COMMAND_EXECUTION]: The transcription script (scripts/transcribe.py) utilizes the subprocess module to execute ffprobe for duration estimation and scripts/json_to_text.py for format conversion. These calls are implemented using argument lists rather than shell strings, which is a standard security practice to prevent shell injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fulfill its primary purpose, including media uploads to Google Cloud Run and Cloud Functions (*.a.run.app, *.cloudfunctions.net), balance checks, and fetching version metadata from the vendor's GitHub repository. These operations target well-known cloud infrastructure services.
[SAFE]: Sensitive information, specifically the TEXTOPS_API_KEY, is managed through a local configuration file (textops_settings.json) or environment variables. The provided settings file contains a placeholder, ensuring no credentials are hardcoded in the skill distribution. Additionally, instructions are provided to the agent to warn users about safety when displaying transcript excerpts.

transcription-speech-to-text-hebrew