The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates on user-supplied content such as article bodies and FAQ sections to generate structured data, which presents an attack surface for indirect prompt injection.
Ingestion points: The process involves reading local files and search results using read_file and search_code to extract entities and breadcrumbs.
Boundary markers: The skill explicitly addresses this risk in its 'Security Guardrails' section (ASI01), directing the agent to reject injection attempts in user-generated blocks.
Capability inventory: Tools like write_file and fetch are available to the agent, representing the impact surface for potential injection.
Sanitization: Instructions recommend machine-validation of JSON-LD outputs and cross-referencing external identifiers with trusted domains like wikidata.org.

implement-schema-markup