The Agent Skills Directory

Command Execution: The skill uses system tools such as the GitHub CLI (gh), ffmpeg, and ffprobe to fetch repository data and process media. These operations are conducted using sanitized inputs (e.g., PR references are normalized and validated) and are confined to the resolved project directory.
External Data Ingestion: As a tool that processes Pull Request descriptions, comments, and code diffs, the skill handles external and potentially untrusted data. It mitigates risk through content cleaning, length budgeting, and the use of bounded 'packets' when delegating work to sub-agents, ensuring that data is treated as information rather than instructions.
Secure Network Communication: Network operations are limited to well-known, trusted domains. The skill fetches contributor avatars exclusively from GitHub's infrastructure with host-validation logic and loads the GSAP animation library from a reputable CDN for use in its generated HTML compositions.
Path Sanitization: The skill implements logic to prevent path traversal, ensuring that all file writes and reads remain within the designated project folder and away from sensitive system locations.
Dynamic Composition Building: The skill assembles HTML and GSAP code at runtime to create video frames. This dynamic generation is part of the core functionality and is based on internal templates and standard libraries, rather than the execution of arbitrary external scripts.

pr-to-video