The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions in SKILL.md utilize shell commands (such as ffprobe and ffmpeg) that interpolate user-provided file paths within double quotes. While quoting provides basic protection against spaces, it does not prevent command injection if the filename contains shell-active characters like backticks or subshells, especially if the agent platform executes these strings directly in a shell environment.
[EXTERNAL_DOWNLOADS]: The skill's setup instructions include installing openai and python-dotenv from the official Python package registry (PyPI). These are trusted, well-known packages commonly used in the industry for API interaction and configuration management.
[INDIRECT_PROMPT_INJECTION]: The skill processes external media files and their metadata, which creates a surface for indirect prompt injection.
Ingestion points: Input file paths and embedded subtitle streams from local video files (referenced in SKILL.md and processed in transcribe.py).
Boundary markers: There are no explicit delimiters or specific instructions for the agent to ignore potentially malicious embedded content within the video metadata or filenames.
Capability inventory: The skill can read local files, write new transcript files to the filesystem, execute system commands via subprocesses, and perform network requests to the OpenAI API.
Sanitization: While the transcribe.py script uses the safer list-based argument format for subprocess.run, the markdown-based instructions for the agent rely on simple string interpolation, which represents a lower degree of sanitization for shell execution.

transcribe-video