The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of user-provided images and text descriptions to generate complex image prompts.
Ingestion points: User-attached images and descriptive text referenced in the task instructions in SKILL.md.
Boundary markers: The instructions include a specific 'Segurança' section explicitly telling the agent to treat text found in images or metadata as data only and to ignore any commands or instruction-overrides found there.
Capability inventory: The skill is limited to text generation; it does not request or use tools for file system writes, network requests, or shell execution.
Sanitization: Instructions require the agent to strictly follow the provided prompt template and ignore external instructions.
[EXTERNAL_DOWNLOADS]: The skill card references the author's GitHub repository for documentation and licensing.
Evidence: https://github.com/marioluciofjr/skills linked in skill-card.md.
Note: This reference points to the official repository of the skill author and represents standard documentation practice.

flip-horizontal