The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as product URLs and image descriptions, which represents a potential attack surface for indirect instructions. Ingestion points: Checkpoint 1 in SKILL.md accepts external product URLs and images for analysis. Boundary markers: The references/research-offers.md file contains a specific safety warning that external pages are untrusted data and not instructions. Capability inventory: The skill has the ability to write to a local Obsidian vault and read from the network to fetch external content. Sanitization: The instructions mandate separating factual evidence from interpretation and treating external data as hypotheses.
[EXTERNAL_DOWNLOADS]: The skill fetches content from external sources for research and inspiration. It includes links to external Notion pages for creative strategies in references/creative-strategy-inspiration.md. It also performs downloads of product images from external URLs into a local directory as defined in references/vault-persistence.md.

morocco-cod-marketing