expo-router

Warn

Audited by Gen Agent Trust Hub on Oct 5, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructs the agent to execute a remote package from the NPM registry using the command npx --yes submit-expo-feedback@latest. This involves downloading and running code at runtime without version pinning or integrity verification.\n- [COMMAND_EXECUTION]: The inclusion of a specific shell command for feedback submission encourages the agent to perform external system operations that may have side effects.\n- [INDIRECT_PROMPT_INJECTION]: The skill implements patterns that ingest untrusted data from URLs and search bars, providing a potential attack surface for indirect prompt injection.\n
  • Ingestion points: useLocalSearchParams hook in references/route-structure.md for URL parameters and headerSearchBarOptions in references/search.md for search input.\n
  • Boundary markers: Implementation examples lack clear delimiters or instructions to treat the ingested data as untrusted.\n
  • Capability inventory: Navigation controls (Link, router.back()) and Stack configuration capabilities.\n
  • Sanitization: The provided code snippets do not include input validation or sanitization before processing search queries or route parameters.\n- [METADATA_POISONING]: The documentation throughout the skill (e.g., in references/tabs.md and references/toolbar-and-headers.md) repeatedly references 'iOS 26' and 'liquid glass' as current platform features. Since these versions do not exist in the current software ecosystem, this information is deceptive and could mislead the agent's generation or the developer's implementation.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 5, 2026, 09:27 PM