The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes user-provided video files using ffprobe and ingests the resulting metadata (duration, resolution, frame rate) into the agent's context to calculate processing parameters. This establishes a surface for indirect prompt injection if the video file contains malicious metadata designed to influence the agent's instructions.
Ingestion points: Video metadata extracted via ffprobe in SKILL.md.
Boundary markers: None used for metadata ingestion.
Capability inventory: Shell command execution (ffmpeg, ffprobe, bash), file system modification, and local server execution.
Sanitization: Video file paths are quoted in shell commands.
[COMMAND_EXECUTION]: The skill performs several automated shell operations to manage files and process media.
Checks for and executes ffmpeg and ffprobe for video frame extraction and analysis.
Orchestrates deployment by executing shell scripts from relative skill directories (e.g., skills/vercel-deploy/scripts/deploy.sh).
Facilitates local testing using npx serve or Python's built-in HTTP server.
[EXTERNAL_DOWNLOADS]: The generated website template incorporates third-party JavaScript libraries via public CDN links.
Fetches the Lenis smooth scroll library from cdn.jsdelivr.net.
Fetches the GSAP animation library and the ScrollTrigger plugin from cdn.jsdelivr.net.

scroll-sequence