The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes ffmpeg for video encoding, scaling, and metadata extraction, and npx playwright for browser-based UI recording. These are essential tools for its primary function of video generation.
[REMOTE_CODE_EXECUTION]: The skill executes a local Node.js script (record-master.js) to manage the recording workflow. It dynamically sets NODE_PATH to resolve modules from sibling 'atoms' (component skills) within the same repository, which is a standard pattern for the platform's modular architecture.
[DATA_EXFILTRATION]: The skill reads project configuration and brand research from the local agent-config/brands/ directory. It uploads finished video files and thumbnails usingauthorized get_upload_url and PUT operations to the platform's storage service. These actions are restricted to the user's project context.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external advertisement templates (recipe.conversation) and user-provided creative 'angles'. While this constitutes an attack surface for indirect prompt injection, the risk is inherent to the skill's purpose of remixing content. The instructions mandate that the AI rewrite the conversation and perform compliance checks to prevent data leakage.

remix-chatgpt-ad-from-sample