The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses spawnSync to invoke system utilities such as ffmpeg, ffprobe, and npm for media processing and dependency management. These commands are executed using array-based arguments rather than shell strings, which effectively mitigates the risk of command injection.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection because it analyzes user-provided SRT files and feedback to generate instructions for sub-agents (e.g., B-roll Director).
Ingestion points: SRT files (srtPath) and user feedback strings collected during approval gates in SKILL.md.
Boundary markers: Absent. The sub-agents receive user data directly as part of their semantic analysis tasks without explicit isolation delimiters.
Capability inventory: The skill can write to project directories, process media via ffmpeg/ffprobe, and install Node.js packages. It lacks capabilities for network exfiltration of sensitive files or accessing system credentials.
Sanitization: Path resolution in scripts/broll-media-utils.js prevents directory traversal by ensuring all asset paths stay within the project's public/ folder. Command arguments are sanitized by the use of spawnSync with argument arrays.
[SAFE]: Dependency management is handled by scripts/ensure-template-deps.js, which restricts automated installation to a hardcoded list of trusted, well-known libraries (remotion, react, lucide-react) within the local project context.

srt-remotion-video-papercraft