The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/transcribe.py utilizes subprocess.run to automate local environment management, including the creation of a virtual environment and the installation of required packages via pip. It also executes ffprobe to determine media duration.
[COMMAND_EXECUTION]: The script scripts/transcribe.py employs os.execv for process replacement, allowing the application to restart itself within the initialized virtual environment to ensure dependency availability.
[EXTERNAL_DOWNLOADS]: The skill automatically downloads and installs several Python packages from public registries and retrieves pre-trained model weights from official OpenAI and HuggingFace repositories.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface in scripts/transcribe.py during the optional punctuation restoration step. 1. Ingestion points: Transcribed text from external media files is passed to the Claude API. 2. Boundary markers: The prompt uses instructions like 'The following is a sentence...' but lacks robust structural delimiters to isolate the transcribed content. 3. Capability inventory: The skill has the ability to execute subprocesses and perform network requests. 4. Sanitization: No sanitization is applied to the transcribed text before it is inserted into the prompt template.

media-transcriber