The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, such as UI plans and existing SwiftUI code, to provide reviews and implementation guidance. This represents a potential surface for indirect prompt injection attacks.
Ingestion points: The "Workflow" and "Review Checklist" sections in SKILL.md direct the agent to process external feature requests and user-provided codebases.
Boundary markers: The instructions do not define delimiters or explicit "ignore embedded instructions" directives to prevent the agent from accidentally executing instructions found within user-provided data.
Capability inventory: The skill is primarily instructional and focused on code generation/review; it does not explicitly invoke high-privilege tools, arbitrary file access, or network exfiltration commands in the provided files.
Sanitization: There is no mention of input validation, escaping, or sanitization protocols for the data being processed.
[EXTERNAL_DOWNLOADS]: The skill references supplementary documentation and API resources hosted on an external, untrusted domain.
Evidence: Multiple markdown links in SKILL.md and references/liquid-glass.md point to https://sosumi.ai/documentation/ for detailed guides on applying and adopting the fictional "Liquid Glass" visual system. While these are currently just informational links, they represent a dependency on a non-standard external source.

swiftui-liquid-glass