The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The file references/cases.md contains numerous links to external media assets and case studies hosted on goodcase.ai and x.com. The skill instructions direct the agent to inspect these finished media items using browser or vision tools to ensure visual fidelity during the prompt generation process.
[INDIRECT_PROMPT_INJECTION]: The skill processes external prompt templates and creator-provided summaries stored in references/cases.md. This represents a surface for indirect prompt injection as the agent ingests content that could theoretically contain instructions. However, the risk is negligible due to the highly structured nature of the workflow.
Ingestion points: The agent reads prompt excerpts and summaries from the references/cases.md file.
Boundary markers: The skill enforces a strict 'Reference Contract' (Preserve / Replace / Avoid) and a verification checklist, which serves as a logical boundary to prevent external content from overriding the agent's primary instructions.
Capability inventory: The skill is limited to producing text-based video prompts and checklists; it does not perform file-system writes, network exfiltration, or command execution.
Sanitization: The instructions explicitly require the agent to 'distinguish observed evidence from your own recommendation', providing a clear separation of data and logic.

product-ad-shot-design