apple-dev-best-practices

Warn

Audited by Gen Agent Trust Hub on Apr 6, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill recommends installing the xcodebuildmcp package from the npm registry using npx -y xcodebuildmcp@latest in references/xcode-claude-integration.md, which downloads unversioned code from an external repository.
  • [REMOTE_CODE_EXECUTION]: The recommended installation method using npx in references/xcode-claude-integration.md involves downloading and executing remote scripts at runtime.
  • [COMMAND_EXECUTION]: The integration guide in references/xcode-claude-integration.md provides configuration for PostToolUse and PreToolUse hooks in .claude/settings.json that automatically execute shell commands (bash, python3, swiftlint) triggered by the agent's file modification activities.
  • [PROMPT_INJECTION]: The skill exhibits keyword stuffing in the description field of SKILL.md, listing numerous Apple frameworks and tools to broaden the skill's trigger surface (Metadata Poisoning).
  • [PROMPT_INJECTION]: The skill processes project files like code and markdown without explicit sanitization or boundary markers, creating a surface for indirect prompt injection.
  • Ingestion points: Processes .swift, .md, and project configuration files from the local environment.
  • Boundary markers: No explicit delimiters or instructions to ignore embedded commands are defined for processed content.
  • Capability inventory: Access to file modification and potentially shell command execution tools.
  • Sanitization: Lacks filtering or validation of content read from external project files.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 6, 2026, 06:14 PM
Security Audit — agent-trust-hub — apple-dev-best-practices