The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/transcribe.py utilizes subprocess.run to orchestrate external command-line tools such as the Claude CLI, GitHub Copilot CLI, and pip for package management. Additionally, examples/basic-transcription.sh executes Python logic through heredocs and shell orchestration.
[INDIRECT_PROMPT_INJECTION]: The skill processes transcription data derived from external audio files, which is subsequently interpolated into prompts for LLMs to generate meeting minutes and summaries. This represents a Category 8 attack surface where speech content could attempt to influence the LLM's output. The workflow uses basic delimiters but lacks deep content sanitization of the transcribed text.
Ingestion points: Local audio files and potential remote URLs (as described in SKILL.md).
Boundary markers: Employs '---' and 'Transcrição:' labels in scripts/transcribe.py to separate instructions from data.
Capability inventory: Includes file system writes, network interactions (via external CLIs), and subprocess orchestration.
Sanitization: Uses safe_user_path to mitigate path traversal risks for output files, though it does not provide deep sanitization of the audio-to-text content itself.
[EXTERNAL_DOWNLOADS]: The scripts/install-requirements.sh and transcribe.py scripts install dependencies from the official PyPI registry and download machine learning models for Whisper transcription.
[PRIVILEGE_ESCALATION]: The skill's documentation and installation scripts include instructions for using sudo to install system-level packages like ffmpeg on Linux environments.
[DYNAMIC_EXECUTION]: The skill performs runtime installation of UI libraries if missing and dynamically generates shell-based orchestrations of Python logic to handle metadata extraction and reporting.

audio-transcriber