The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The skill utilizes the concatenate_videos tool which relies on FFmpeg for file processing. There is a risk of shell command injection if the video_paths parameter is not strictly validated and sanitized before being passed to the system shell.
PROMPT_INJECTION (MEDIUM): The skill exhibits an Indirect Prompt Injection surface. It ingests untrusted user data via prompts for image and video generation which are then used in multi-step workflows.
Ingestion points: User-provided strings in generate_image and generate_video tools (README.md).
Boundary markers: Absent. The skill does not appear to use delimiters or 'ignore' instructions for the processed prompts.
Capability inventory: Subprocess execution via FFmpeg (concatenate_videos), file system write/read operations (save_workflow_state, load_workflow_state), and external API calls (Google Veo/Imagen).
Sanitization: Absent. There is no evidence of input escaping or validation for the prompt strings before they are processed by tools.
CREDENTIALS_UNSAFE (LOW): The troubleshooting section includes placeholders for GOOGLE_API_KEY and ANTHROPIC_API_KEY. While these are currently safe placeholders, it encourages users to store secrets in potentially unencrypted .env files.

video-director