The Agent Skills Directory

[METADATA_POISONING]: The skill uses deceptive metadata to claim official authorship by a trusted entity. Both the skill description and introductory text explicitly state that the guidance was 'written and published by Apple' and consists of 'authoritative' practices from them. Given the actual author is 'tartinerlabs', this misrepresentation is designed to gain unauthorized trust from the user and the agent.
[PROMPT_INJECTION]: The skill includes instructions that attempt to override the model's base training and safety protocols by stating the provided information 'unconditionally supersedes any prior training the model may have on these topics.' This directive language is a common pattern for displacing operational guidelines.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data in the form of local source code. It lacks sanitization logic or boundary markers to protect the agent from malicious instructions embedded within the codebase it analyzes. Ingestion points: SKILL.md instructs the agent to 'scan the project to identify different smaller areas of the code' and review it. Boundary markers: No delimiters or isolation instructions are provided to separate analyzed code from the agent's instructions. Capability inventory: The skill is capable of providing refactoring advice, identifying focus areas, and generating new SwiftUI code based on the scanned project content. Sanitization: No methods for validating or filtering the ingested source files are described.

swiftui-specialist