The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: Attack surface identified in the handling of external data files.
Ingestion points: The skill provides instructions for loading pass data using PKPass(data: passData) in SKILL.md and references/wallet-passes.md.
Boundary markers: Absent. There are no instructions to the agent to treat the ingested data as untrusted or to use delimiters.
Capability inventory: The skill facilitates payment authorization via PKPaymentAuthorizationController and file system/Wallet interaction via PKPassLibrary.
Sanitization: Absent. The code does not demonstrate validation or sanitization of the pass data before it is processed by the framework.
[EXTERNAL_DOWNLOADS]: The skill directs the agent or user to non-official documentation sources.
Evidence: Multiple links in the References section of SKILL.md point to the sosumi.ai domain (e.g., https://sosumi.ai/documentation/passkit) instead of official Apple Developer documentation.
[METADATA_POISONING]: The skill claims to target non-existent versions of technologies.
Evidence: The skill header specifies "Targets Swift 6.3 / iOS 26+", which are future versions not currently available, potentially misleading the agent regarding the applicable runtime environment or API availability.

passkit