The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub Pull Requests, including titles, descriptions, and code diffs. \n
Ingestion points: Data is fetched via gh and transformed into a narrative source file (visible-text.txt) by ingest.mjs. \n
Boundary markers: The frame-packets.mjs script builds isolated data packets to limit the context provided to frame-worker sub-agents. \n
Capability inventory: The skill utilizes gh, ffmpeg, and node to manipulate the local file system and generate project assets. \n
Sanitization: ingest.mjs includes a routine to strip HTML comments from the PR body to prevent simple injection techniques. \n- [COMMAND_EXECUTION]: The workflow relies on several external command-line tools. \n
Executes gh (GitHub CLI) to retrieve pull request details, diffs, and contributor information. \n
Spawns ffmpeg and ffprobe processes to analyze and manipulate media durations. \n
Invokes node to run internal helper scripts for project initialization, design remixing, and assembly. \n- [EXTERNAL_DOWNLOADS]: Fetches external resources to populate the video project. \n
Downloads contributor avatars from GitHub; fetch-people-avatars.mjs implements a host-based whitelist (SSRF guard) to restrict downloads to GitHub-controlled domains. \n
Retrieves audio and voice assets from the HeyGen API using configured credentials. \n- [DYNAMIC_EXECUTION]: Assembles executable content at runtime for animation. \n
Scripts like captions.mjs and transitions.mjs inject GSAP (GreenSock Animation Platform) code into the final index.html file. \n
Frame workers generate HTML and CSS fragments based on design presets to create individual video compositions.

hyperframes-pr-to-video