The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from website URLs which may contain malicious instructions designed to influence the agent's behavior during the video generation process.
Ingestion points: Website URLs and local project directory contents (SKILL.md).
Boundary markers: No specific delimiters or safety instructions are mentioned to isolate external website content from the agent's instructions.
Capability inventory: The skill utilizes FFmpeg for audio/video processing, browser automation (Playwright/Puppeteer) for frame capture, and has file-writing capabilities to the brag-output/ directory.
Sanitization: The skill does not describe any sanitization or filtering of content fetched from external URLs before processing.
[EXTERNAL_DOWNLOADS]: The skill is a 'catalog stub' that fetches its operational instructions and logic from an external, non-whitelisted GitHub repository.
Evidence: References the upstream repository at https://github.com/latent-spaces/brag and indicates that the installation process pulls the current file live from that location.
[DYNAMIC_EXECUTION]: The skill relies on the language model to dynamically construct the execution plan for video generation, including storyboarding, visuals, and sound mixing.
Evidence: The description states 'The model builds the whole launch video itself (story, visuals, music, sound effects, mix and render) with the tools already on the machine'.
[COMMAND_EXECUTION]: The skill executes shell commands and scripts through local tools to perform its primary function.
Evidence: Uses FFmpeg for encoding and headless browsers (Node with Playwright, Puppeteer, or Chrome) to render HTML frames and capture website assets.

brag-slim