The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for processing external audio data from URLs and local files. This creates a surface for indirect prompt injection where the transcribed text might contain instructions for an LLM that processes the output.\n- Ingestion points: audioUrl parameter in analyzeConversation (SKILL.md) and path parameters in transcription functions (SKILL.md and references/code-patterns.md).\n- Boundary markers: Not present in these SDK implementation examples.\n- Capability inventory: The code utilizes network access via the Deepgram SDK and file system operations to read and write audio files.\n- Sanitization: Input validation for audio sources and sanitization of transcript strings are not included in the provided snippets.\n- [SAFE]: The skill utilizes official Deepgram SDKs and recommends using environment variables for API key management, which is a secure practice.

deepgram-sdk-patterns