The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for fetching and processing data from external network sources via URLSession. This capability creates an attack surface for indirect prompt injection if untrusted API responses are subsequently used to influence agent decisions.
Ingestion points: Network requests in SKILL.md, references/urlsession-patterns.md, and references/background-websocket.md ingest data via URLSession.data, URLSession.bytes, and URLSession.download tasks.
Boundary markers: Absent; the skill does not define specific delimiters or instructions to isolate or treat external data as untrusted within the agent's context.
Capability inventory: The skill enables network operations (REST, WebSockets), file system modifications (moving downloads), and background task execution.
Sanitization: The skill demonstrates the use of JSONDecoder for structured data validation, though it lacks explicit sanitization for free-text content before use in prompts.
[METADATA_POISONING]: The skill uses futuristic or illustrative version numbers (iOS 26+, Swift 6.3) and references an external domain (sosumi.ai) for documentation links. While these are potentially misleading regarding current platform capabilities, they do not appear to be malicious or intended for deception that compromises safety.

ios-networking