The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to visit and process content from external website URLs provided by the user to 'read your site' and 'write the story'. This creates a surface where a malicious website could attempt to influence the agent's behavior.
Ingestion points: Untrusted content is ingested from website URLs passed as the primary input for the video generation process (e.g., in README.md and SKILL.md).
Boundary markers: The instructions do not define specific delimiters or instructions to ignore embedded prompts in the scraped website content.
Capability inventory: The agent has access to powerful tools including playwright for browser automation, ffmpeg for video manipulation, and the ability to execute node and python scripts. It is also instructed to use the serve tool to share the resulting media files.
Sanitization: No specific sanitization or filtering of external content is mentioned before it is processed by the LLM to generate the video brief.
[COMMAND_EXECUTION]: The skill relies on and instructs the agent to use various CLI tools and local scripts for video assembly, sound synthesis, and mastering.
Evidence: The legacy/scripts folder contains several .mjs and .sh files used to wrap ffmpeg, ffprobe, and node operations. SKILL.md also specifies the use of playwright and python3 for beat analysis.
[DATA_EXPOSURE]: The skill instructs the agent to access sensitive local data to enhance the authenticity of the product videos.
Evidence: SKILL.md contains a 'Non-negotiable' rule to 'Use real data', specifically prompting the agent to 'Pull real, defensible numbers read-only (creators reached, gifts sent, posts/views tracked)' from the user's database to display on rolling odometers in the video.

product-launch-motion