The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from an external reference file, which represents a potential ingestion point for untrusted instructions, although the content currently consists only of image prompt metadata.
Ingestion points: The workflow requires the agent to read references/cases.md at the start of each invocation to select an anchor case.
Boundary markers: A logical boundary is established through a mandatory "Preserve / Replace / Avoid" reference contract and a verification checklist that verifies consistency between the evidence and the generated artifact.
Capability inventory: The skill is designed to generate textual image prompts and may use browser or image inspection tools to verify visual evidence.
Sanitization: The skill relies on structured output requirements and comparative logic to manage the influence of external data rather than programmatic sanitization.
[EXTERNAL_DOWNLOADS]: The skill provides links to external URLs for visual evidence and creator attribution.
Evidence: References are provided to goodcase.ai, media.goodcase.ai, xiaohongshu.com, x.com, and douyin.com. These links allow the agent or user to inspect the "finished media" associated with each case study. These represent vendor-owned resources and well-known content platforms, and their use is consistent with the skill's primary function.

editorial-portrait-composition