The Agent Skills Directory

[COMMAND_EXECUTION]: The SKILL.md file executes a Python script using the command python3 ~/.claude/skills/video-analyzer/scripts/analyze_video.py $ARGUMENTS. This direct interpolation of $ARGUMENTS into a shell execution block without sanitization or escaping creates a potential command injection vulnerability. An attacker or a malicious user input containing shell metacharacters (e.g., ;, &&, |) could execute arbitrary commands on the host system.
[PROMPT_INJECTION]: The skill processes untrusted external data from video files and accepts a user-provided --prompt argument, both of which are interpolated into the prompt sent to the Gemini model. This creates an indirect prompt injection surface (Category 8) where malicious instructions embedded in the video or the custom prompt could influence the agent's behavior.
Ingestion points: The skill ingests data from local video files (video_path) and a user-provided prompt string (--prompt via $ARGUMENTS).
Boundary markers: No delimiters or boundary markers are used to separate user-provided arguments from the shell command, and no 'ignore embedded instructions' warnings are present for the model when processing video content.
Capability inventory: The skill is granted access to the Bash and Read tools, allowing it to execute system commands and read files from the disk.
Sanitization: There is no evidence of sanitization for shell metacharacters in the argument processing, nor is there filtering or validation for the content being sent to the AI model.

video-analyzer