The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/transcribe.py executes external command-line tools claude and gh copilot using subprocess.run. These calls are used to process the transcription text into structured documentation and are an intended part of the skill's functionality. Additionally, the script automatically attempts to install missing dependencies (rich, tqdm) using pip at runtime if they are not detected.
[EXTERNAL_DOWNLOADS]: The scripts/install-requirements.sh script installs necessary Python dependencies via pip and downloads a Whisper transcription model from a well-known registry (HuggingFace) during the initialization process. It also facilitates the installation of system utilities like ffmpeg via Homebrew on macOS.
[PROMPT_INJECTION]: The skill processes untrusted content from audio files and includes it in prompts sent to LLMs, creating a surface for indirect prompt injection.
Ingestion points: Transcribed text from user-provided audio files is captured in scripts/transcribe.py and interpolated into prompt templates.
Boundary markers: The script uses a simple triple-dash (---) separator as a delimiter between the system instructions and the untrusted transcription data.
Capability inventory: The skill possesses the capability to invoke external CLI tools (claude, gh) and write Markdown files to the local file system.
Sanitization: No specific input sanitization, filtering, or escaping is applied to the transcribed text before it is embedded into the LLM prompt.

audio-transcriber