The Agent Skills Directory

Authentication and Credential Management: The skill provides code templates for handling API keys and OAuth sessions. These are standard implementation patterns for authenticated network services. Developers utilizing these templates should ensure that keys are stored securely (e.g., using Keychain as suggested in the text) and that the Hashable implementation for configurations does not inadvertently expose sensitive data in logs.
External Data Processing (Attack Surface): The executor component acts as a bridge for data received from remote inference APIs. This creates a surface for indirect prompt injection if the remote model produces malicious content.
Ingestion points: The respond method in the Executor implementation (Step 2) ingests data from a remote provider via the openProviderStream sequence.
Boundary markers: The code focus is on protocol translation; it does not explicitly implement data delimiters, as sanitization is typically handled by the underlying Foundation Models framework or the inference provider.
Capability inventory: The executor performs network operations to communicate with external endpoints and uses the LanguageModelExecutorGenerationChannel to stream data back to the agent framework.
Sanitization: The skill provides a structural template for translation. Developers implementing this should include validation logic within the init(configuration:) and response handling blocks to verify the integrity of external data.

foundation-models-language-model-protocol