The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves collecting user-provided inputs (subject, setting, etc.) and interpolating them into a structured prompt template grounded in examples from references/cases.md.
Ingestion points: User input for video subject, setting, and duration (SKILL.md, Workflow step 1).
Boundary markers: The workflow uses a structured block-by-block approach, but there are no explicit delimiters or instructions to ignore potential commands embedded in the user-supplied data.
Capability inventory: The skill generates text-based video prompts. It does not possess file-write, network, or subprocess capabilities.
Sanitization: There is no evidence of input validation or sanitization before the user data is incorporated into the final prompt.
[EXTERNAL_DOWNLOADS]: The skill and its reference file contain numerous links to external domains that are not part of the trusted vendor list.
Evidence: references/cases.md contains links to goodcase.ai, x.com, and t.co shortened URLs.
Risk: While these links are provided as part of a dataset for 'anchor cases', they lead to unverified external content.
[DATA_EXPOSURE]: Case E1 in references/cases.md contains a raw referral/affiliate link (https://t.co/l1lPYN3ENF) captured from a social media post.
Evidence: The text in E1 includes: "Use my referral link to get offer: https://t.co/l1lPYN3ENF".
Risk: An agent instructed to "ground the draft in one anchor case" or to "follow the structure" may inadvertently propagate these marketing links or shortened URLs into the user's final video prompt, which could be used for spam or affiliate fraud.

seedance-pet-animal