ios-networking

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill implements remote data fetching patterns that could expose an agent to malicious instructions embedded in external API responses.
  • Ingestion points: Data is ingested via URLSession.shared.data(for:), session.downloadTask, and session.bytes(for:) across SKILL.md, references/urlsession-patterns.md, and references/background-websocket.md.
  • Boundary markers: The implementation includes HTTP status code validation and schema-based JSON decoding (Codable), but lacks explicit sanitization or delimiters for data that might be passed to a large language model's context.
  • Capability inventory: The skill includes extensive capabilities for network communication (URLSession, Network.framework) and local file system manipulation (FileManager), including background transfers and file protection management.
  • Sanitization: The patterns rely on JSONDecoder for structured data enforcement, which mitigates schema confusion but does not address potential prompt injection content within string values.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 03:42 PM
Security Audit — agent-trust-hub — ios-networking