The Agent Skills Directory

[SAFE]: The skill serves as a diagnostic and educational resource for the Foundation Models framework. All code snippets provided are standard Swift implementations for local AI model interaction on Apple platforms.
[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for processing user-supplied data through LLM sessions (e.g., session.respond(to: userInput)). While this identifies a potential attack surface where untrusted data could influence model behavior, the skill specifically guides developers on how to handle safety-related framework responses.
Ingestion points: The skill demonstrates passing userInput to the LanguageModelSession.respond method in SKILL.md (Patterns 2b, 2c, and 3c).
Boundary markers: The snippets focus on error handling rather than prompt construction; explicit delimiters for user input are not demonstrated.
Capability inventory: The code patterns illustrate text generation, structured data parsing using the @Generable macro, and external integration via the Tool protocol.
Sanitization: The skill highlights best practices for safety, such as catching guardrailViolation errors and advising against storing potentially harmful user input in logs.

axiom-foundation-models-diag