The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the agent to access and "inspect" finished media (videos and images) from external URLs hosted on well-known services like x.com (Twitter) and xiaohongshu.com, as well as goodcase.ai. This is an intended function of the skill used to verify visual styles for prompt engineering.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from referenced URLs and prompt excerpts, which creates a theoretical attack surface for indirect prompt injection if the external sites were to contain malicious instructions.
Ingestion points: Reference cases in references/cases.md containing external links and user-supplied descriptions.
Boundary markers: The workflow mandates the creation of a "reference contract" using Preserve, Replace, and Avoid categories to explicitly filter and define which traits are adopted from the evidence.
Capability inventory: The skill utilizes image and video inspection tools or browser capabilities to analyze external media.
Sanitization: The instructions require the agent to distinguish observed evidence from its own recommendations and labels results as "prompt-grounded" if media is inaccessible.

epic-fantasy-scifi-spectacle