The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent on how to ingest data from the device's contacts database, establishing a surface for indirect prompt injection if contact fields contain malicious payloads.
Ingestion points: Contact data is enumerated and fetched using CNContactFetchRequest and enumerateContacts within SKILL.md and references/contacts-patterns.md.
Boundary markers: The skill does not provide instructions for using delimiters or boundary warnings to protect the agent from instructions embedded in contact data fields (e.g., names or notes).
Capability inventory: The skill enables high-privilege operations such as modifying contact records (CNSaveRequest), exporting data to vCards (CNContactVCardSerialization), and managing contact groups.
Sanitization: No input validation or escaping protocols are provided for the data retrieved from external contact records.
[EXTERNAL_DOWNLOADS]: The documentation references include links to an unofficial third-party domain instead of official Apple developer resources.
Evidence: Multiple references in SKILL.md point to sosumi.ai for framework documentation and API specifications.

contacts-framework