The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates the execution of ffmpeg and ffprobe by constructing shell commands that incorporate user-provided file paths and settings. The instructions utilize quoting (e.g., "$INPUT_FILE") to mitigate basic command injection, though security depends on the agent's ability to safely handle shell metacharacters in filenames.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by reading video metadata via ffprobe JSON output.
Ingestion points: Stream and format metadata from user-supplied video files (SKILL.md).
Boundary markers: The skill uses structured JSON parsing but lacks explicit delimiters or instructions to ignore instructions potentially hidden in metadata tags.
Capability inventory: Shell execution of ffprobe and ffmpeg, and file writing capabilities via cat (SKILL.md).
Sanitization: There are no explicit instructions to sanitize or filter metadata fields before they are displayed to the user or used in command construction.
[DYNAMIC_EXECUTION]: The skill dynamically generates temporary configuration files (/tmp/concat_list.txt) and complex shell command strings at runtime based on the analysis of input files and user-specified preferences.

video-concat