The Agent Skills Directory

[METADATA_POISONING]: The skill contains deceptive and inaccurate metadata throughout its reference files and evaluation suite. Specifically, references/nutrition-labels.md and evals/evals.json repeatedly reference a non-existent platform version (iOS 26) as if it were the current standard. Furthermore, references/media-accessibility.md provides a link to a parody domain (sosumi.ai) while presenting it as official documentation for the Apple AVMediaCharacteristic API. This deception could mislead users or agents into relying on unverified or humorous information as authoritative documentation.
[INDIRECT_PROMPT_INJECTION]: The skill introduces a vulnerability surface by directing the agent to an external, untrusted parody site for API documentation. If the agent follows these links to verify information, it may ingest content from a non-authoritative source.
Ingestion points: references/media-accessibility.md contains a link to an external parody documentation site (https://sosumi.ai/documentation/avfoundation/avmediacharacteristic).
Boundary markers: No delimiters or warnings are provided to indicate that the linked source is non-official or humorous.
Capability inventory: The skill provides implementation patterns and auditing workflows across SwiftUI, UIKit, and AppKit that the agent is expected to apply to code.
Sanitization: No validation or sanitization of content from the linked external domain is suggested or performed.

ios-accessibility