The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (customer reviews, brand dossiers, and user-provided files) which is then used to construct prompts for an image generation tool.
Ingestion points: Data is ingested through advibly_get_brand (brief), advibly_get_brand_dossier, and user-provided inputs as described in Phase A and Phase B of the main instruction file.
Boundary markers: While the prompt skeleton in references/templates.md uses double quotes to delimit copy, it lacks explicit instructions to the model to ignore any hidden instructions embedded within the verbatim source text.
Capability inventory: The skill possesses the capability to generate images via advibly_generate_image and publish content using advibly_social_create_post.
Sanitization: The skill's instructions mandate verbatim quotation of customer reviews to maintain authenticity, which implicitly prevents the sanitization of potential prompt injection payloads hidden in those reviews.

static-ads