The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the ffmpeg binary to perform video transformations.
Evidence: In scripts/vfilter.py, the script uses subprocess.call(cmd) where cmd is a list of command-line arguments.
Security Context: The use of argument lists instead of raw strings mitigates standard shell injection vulnerabilities. However, the script's --custom mode allows arbitrary filter strings to be passed to ffmpeg, which is a powerful engine capable of file system interactions.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it processes untrusted data and translates it into powerful tool operations.
Ingestion points: The skill accepts user-provided arguments through the $ARGUMENTS placeholder in SKILL.md and CLI flags in scripts/vfilter.py, such as --filter-string and --input.
Capability inventory: The skill can read and write files on the local system via ffmpeg. Certain ffmpeg filters (like drawtext using the textfile option) can be used to disclose the contents of local files.
Boundary markers: The skill does not implement boundary markers or instructions to warn the agent about potentially malicious instructions embedded in the data it processes.
Sanitization: No validation or sanitization is performed on the input file paths or the custom filter strings, relying on the agent's internal guardrails to prevent misuse.

ffmpeg-video-filter