The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as screenshots, recordings, and code diffs to generate video content, creating a surface for potential indirect prompt injection attacks.
1. Ingestion points: Work artifacts including screenshots, recordings, and diffs (SKILL.md).
1. Boundary markers: No specific delimiters or warnings for the agent to ignore embedded instructions are present.
1. Capability inventory: Rendering video files (file-write) and likely command execution via media tools (SKILL.md).
1. Sanitization: No evidence of data sanitization or validation of the ingested artifacts.
[EXTERNAL_DOWNLOADS]: The skill references source code and documentation from a well-known repository (github.com/cursor/plugins). These references are documented neutrally.
[COMMAND_EXECUTION]: The skill requires the agent to render video files (e.g., mp4), which involves executing media processing tools in the agent's environment.

ketchup-video