mapkit

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied search queries through autocomplete and search APIs. \n
  • Ingestion points: The query fragment in SearchCompleter (SKILL.md) and searchText in MapSearchView (references/mapkit-patterns.md).\n
  • Boundary markers: Not explicitly implemented for the search strings.\n
  • Capability inventory: Utilizes MKLocalSearch and MKLocalSearchCompleter for network-based location lookups.\n
  • Sanitization: Includes debouncing (300ms) to throttle input processing.\n- [METADATA_POISONING]: The skill includes documentation links targeting 'sosumi.ai' and refers to a non-existent 'iOS 26' version. These items appear to be placeholders or illustrative examples rather than attempts to deceive the agent for malicious purposes.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 03:42 PM
Security Audit — agent-trust-hub — mapkit