swift-macos

Pass

Audited by Gen Agent Trust Hub on Mar 19, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references downloading and installing the swiftly toolchain manager from swift.org and various dependencies through the Swift Package Manager. These are well-known and official sources for Swift development.
  • [COMMAND_EXECUTION]: Provides instructions and examples for using command-line tools such as xcodebuild, notarytool, and stapler for application distribution, as well as tccutil for managing system permissions during development. It also includes a local validation script in package.json.
  • [DATA_EXFILTRATION]: Documents the use of ScreenCaptureKit and AVFoundation for capturing screen content, application audio, and microphone input. While these are core features of the documented framework, they represent a sensitive data access surface.
  • [PROMPT_INJECTION]: Documents the Foundation Models framework for on-device AI, which creates an indirect prompt injection surface as it teaches building apps that process LLM-generated content.
  • Ingestion points: references/foundation-models.md uses LanguageModelSession.respond(to:) to process arbitrary text.
  • Boundary markers: Absent in the provided code snippets.
  • Capability inventory: The skill covers extensive system capabilities including file system access (references/system-integration.md) and shell command execution (references/distribution.md).
  • Sanitization: No explicit sanitization or validation of LLM output is demonstrated in the examples.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 19, 2026, 11:43 AM