The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The file references/cases.md contains multiple links to external domains including goodcase.ai, media.goodcase.ai, x.com, and the shortened URL t.co/FMs1prnPP0. These are used to provide the agent with examples of finished media and prompt structures.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read and follow 'Case evidence' in references/cases.md, which contains prompt excerpts. These excerpts include high-authority instructions such as 'its rules... override your defaults' and directives to 'clone and read' external repositories. This creates a surface where the agent could adopt instructions from untrusted or compromised external sources.
Ingestion points: The agent is mandated to read references/cases.md and inspect the external media links contained within on every invocation.
Boundary markers: The skill attempts to mitigate this by requiring a 'Preserve/Replace/Avoid' reference contract to filter what logic is transferred.
Capability inventory: The agent is expected to use browser or media inspection tools and produce complex artifacts like HTML/JS code or video prompts.
Sanitization: The workflow includes a verification step to compare the artifact with the anchor case and check for continuity risks.
[DYNAMIC_EXECUTION]: The workflow requires the agent to 'Turn text, data, and visual elements into controllable parameters' and produce a 'single file that can be screen-recorded as is,' typically an HTML or JavaScript file. This involves the generation of code intended for immediate rendering and execution by the user or a browser tool.

code-rendered-explainer-video