The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources (Apple documentation and WWDC transcripts) to update the repository. Although the sources are intended to be official, the workflow lacks boundary markers or sanitization for the ingested content, creating a theoretical attack surface. * Ingestion points: fetchAppleDocumentation and fetchExternalDocumentation tools. * Boundary markers: None present in instructions. * Capability inventory: The skill uses gh pr create and has write access to repository files. * Sanitization: No validation of the fetched documentation content is mentioned.
[METADATA_POISONING]: The references/scan-manifest.md file includes sections for fictional software versions (iOS 26+) and future events (WWDC 2025), which provides inaccurate context to the agent.

update-swiftui-apis