The Agent Skills Directory

[SAFE]: The skill provides architectural patterns and UI components for SwiftUI development. Analysis of the code snippets indicates legitimate use of Apple's Speech and AVAudioSession APIs for transcription and audio level monitoring.
[DATA_EXPOSURE]: The implementation correctly manages microphone and speech recognition permissions, ensuring that users are prompted for access and that the application handles denial gracefully by disabling the listening UI and providing a link to system settings.
[PROMPT_INJECTION]: The skill facilitates the ingestion of user-controlled voice transcripts that are subsequently processed by an agent.
Ingestion points: Untrusted voice input is captured via the SpeechRecognizer.transcript property in references/speech-engine.md and handed to the agent via ChatRequest.initialSend in references/agent-wiring.md.
Boundary markers: The code snippets do not illustrate the use of specific boundary markers or delimiters when interpolating the transcript into agent prompts.
Capability inventory: The skill utilizes network-based agent communication to process transcripts and maintains a sessionThreadId for conversation history.
Sanitization: The implementation performs basic whitespace trimming on the transcript but does not include explicit escaping or validation logic for the content being sent to the LLM.
[COMMAND_EXECUTION]: No unauthorized or dangerous command execution patterns were found. Technical advice regarding platform-specific APIs (e.g., AssetInventory) is provided for application stability.

swiftui-voice-first-agent-chat