The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/frame.sh is vulnerable to command injection via the --index, --time, and --out parameters. These values are interpolated directly into shell commands and ffmpeg filtergraphs without sufficient sanitization, which could allow an attacker to execute arbitrary shell commands or manipulate video processing filters. \n
Ingestion points: Command-line arguments processed in scripts/frame.sh. \n
Boundary markers: None; the parameters are used directly within shell strings and subshells. \n
Capability inventory: The script executes ffmpeg and mkdir with subshell calls to dirname. \n
Sanitization: Incomplete; while variables are double-quoted in some places, the script does not prevent subshell expansion (e.g., within the --out path) or escaping of the ffmpeg filtergraph syntax. \n- [PROMPT_INJECTION]: The SKILL.md file contains a commercial pricing table and contact information for paid video services. This is a form of metadata poisoning that is irrelevant to the skill's technical purpose and serves as promotional clutter for the agent.

video-frames