The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill implements a workflow where the agent is prompted to dynamically generate Remotion (React/TypeScript) code based on a 'style brief' extracted from untrusted external reference videos. This generated code is subsequently executed via the npx remotion render command.
[EXTERNAL_DOWNLOADS]: The skill instructions frequently involve downloading content and executable extensions from third-party sources and unverifiable GitHub repositories, such as d.fxtwitter.com, mcp-remote URLs (e.g., https://mcp.tweetsave.org/sse), and various individual GitHub accounts for MCP servers (e.g., lurenss/linkedin-mcp).
[COMMAND_EXECUTION]: The pipeline relies heavily on the execution of powerful system commands including brew install, npx, curl, and git clone. It specifically guides the user or agent to install and run numerous third-party CLI tools and MCP servers to handle social media interactions.
[PROMPT_INJECTION]: The skill has a significant indirect prompt injection surface. It is designed to ingest untrusted data from external URLs (X/Twitter, YouTube, etc.) and pass it to an LLM for analysis (Phase 0). The output of this analysis ('style brief') then controls the logic and content of the entire downstream generation pipeline (Narrative, Visuals, and Video phases).
Ingestion points: SKILL.md and references/x-content-extraction.md via curl and yt-dlp from user-supplied URLs.
Boundary markers: Limited; instructions primarily rely on requesting structured JSON output from the LLM.
Capability inventory: Includes exec, curl, bun add, git clone, ffmpeg, and npx remotion render for dynamic code execution.
Sanitization: No explicit sanitization or filtering of the content extracted from reference posts is mentioned before it influences downstream code generation.

content-creation