The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The script scripts/frame.sh constructs ffmpeg commands using unvalidated input variables $time and $index.
Evidence: In scripts/frame.sh, the ${index} variable is interpolated directly into the -vf (video filter) argument: "select=eq(n\\,${index})". If an agent populates this parameter with unvalidated user input, an attacker could inject additional ffmpeg filters (e.g., using a value like 0),drawtext=...) to manipulate output or potentially exploit tool-specific vulnerabilities.
[PROMPT_INJECTION] (MEDIUM): The skill is susceptible to Indirect Prompt Injection (Category 8) due to its processing of untrusted external data.
Ingestion points: The script scripts/frame.sh ingests external video files provided via the <video-file> argument.
Boundary markers: None. The script does not use delimiters or instructions to prevent the agent from being influenced by data found within the video metadata or subtitles during processing.
Capability inventory: The skill executes ffmpeg, a powerful media processing tool capable of complex operations, including reading local files and making network requests (e.g., via HLS playlists or the concat demuxer) if the input file is malicious.
Sanitization: The script only checks for file existence ([[ ! -f "$in" ]]) and does not validate the integrity or content of the video file before processing.

video-frames