The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a bash script (extract-frames.sh) to run ffmpeg and ffprobe on user-provided video paths. Interpolating user-controlled file paths into shell commands without explicit sanitization instructions creates a potential command injection surface.
[PROMPT_INJECTION]: The skill ingests untrusted data from video files (metadata and audio transcripts) to generate prompt templates. 1. Ingestion points: Video metadata (duration, resolution) and audio content (transcribed via Whisper) from user-supplied files. 2. Boundary markers: None identified; there are no delimiters or instructions to ignore malicious content within the analyzed transcripts. 3. Capability inventory: Executes local shell scripts, writes new files to the prompting library, and performs network operations via the Novoads API. 4. Sanitization: No validation is performed on the transcripts or metadata before they are incorporated into generated template files.
[EXTERNAL_DOWNLOADS]: Recommends installation of ffmpeg (via Homebrew) and openai-whisper (via pip). These are standard, well-known, and reputable tools.
[DATA_EXFILTRATION]: Communicates with the vendor domain api.novoads.ai to upload assets, get estimates, and generate videos. These operations are consistent with the skill's primary purpose and use the author's official infrastructure.

analyze-video