The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes external video files provided by the user, creating a surface for indirect prompt injection.
Ingestion points: Video files are processed to extract metadata, frames, and audio transcripts as described in SKILL.md.
Boundary markers: While the skill provides structural templates for the output, it lacks explicit delimiters or warnings to ignore instructions that might be embedded in analyzed data (e.g., spoken dialogue or text-on-screen).
Capability inventory: The skill can execute local shell commands (via extract-frames.sh), write files to the local prompt library, and make authenticated API calls to the Novoads platform.
Sanitization: No explicit sanitization or filtering of transcribed dialogue is mentioned before it is used to generate new prompting templates.
[COMMAND_EXECUTION]: The skill uses local command-line tools, including ffmpeg, ffprobe, and whisper, along with a shipped shell script extract-frames.sh. These tools are used for frame extraction and transcription, which are primary functions of the skill and run locally on the user's files.
[EXTERNAL_DOWNLOADS]: The skill documentation guides the user to install standard third-party tools such as ffmpeg and the openai-whisper Python package. These are reputable dependencies required for media processing.
[SAFE]: The skill follows secure secret management practices by instructing users to provide their API key via environment variables (.env files) rather than allowing hardcoded credentials. All network activity is directed to the official vendor domain (novoads.ai).

analyze-video