The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script render_demo.py executes system binaries such as vhs, ffmpeg, and ffprobe. These calls are implemented using list-based arguments without shell=True, which is a secure practice that prevents shell injection.
Evidence: The script uses subprocess.run(command, ...) where command is a list of strings, effectively bypassing shell interpretation.
[INDIRECT_PROMPT_INJECTION]: The skill processes .tape files to identify output paths and validate their location. While this involves reading user-controlled data, the script implements strict path validation to ensure files stay within the repository's resource directories.
Evidence: Path validation is performed using tape.relative_to(resources_root) and shlex.split is used for safe command parsing of the tape file contents.
[SAFE]: The skill instructions emphasize security by explicitly forbidding the inclusion of real credentials or production data in the terminal recordings.
Evidence: SKILL.md includes the directive: 'Never put real credentials, tokens, personal paths, or production data in a tape or fixture. Use unmistakably fake values.'

record-vhs-demo