The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the package expo-apple-intelligence using npx expo install. This package is not part of the official @expo scope, and its provenance is unverified. Recommending unverified third-party packages poses a supply chain risk, as an attacker could register a package with this name to execute malicious code during installation.
[METADATA_POISONING]: The skill contains misleading or hallucinatory technical claims, such as requirement for 'iOS 26+' (a version significantly beyond the current release cycle) and 'Apple Foundation Models' availability in this context. This may lead the AI agent to provide inaccurate or invalid advice to developers.
[INDIRECT_PROMPT_INJECTION]: The skill provides code examples that ingest untrusted user prompts into an LLM session without implementing security boundaries or sanitization.
Ingestion points: session.respond in SKILL.md accepts a prompt string.
Boundary markers: Absent in the provided code snippets.
Capability inventory: Access to on-device LLM generation and structured output via the expo-apple-intelligence library.
Sanitization: No input validation, escaping, or filtering logic is suggested or implemented.

add-foundation-models