The Agent Skills Directory

Indirect Prompt Injection Surface: Several patterns in the skill involve ingesting untrusted data and processing it through Large Language Models (LLMs) for tasks like sentiment analysis, summarization, or chatbot responses.
Ingestion points: The skill retrieves meeting transcripts, summaries, and chat messages from Zoom APIs and real-time streams (e.g., in use-cases/ai-integration.md and use-cases/team-chat-llm-bot.md).
Boundary markers: The provided code snippets demonstrate direct interpolation of untrusted content into LLM prompts without explicit delimiters or instructions to the model to ignore potentially embedded commands.
Capability inventory: The demonstrated workflows involve network operations to Zoom APIs and external AI service providers (e.g., OpenAI, Anthropic, AssemblyAI).
Sanitization: Explicit input sanitization or filtering logic is not present in the reference implementations, which is a potential consideration for developers following these patterns.
External Service Integrations: The skill references and downloads resources from established technology providers and official Zoom infrastructure.
Well-known services: It provides examples using Amazon Web Services (S3), Redis, OpenAI, and Anthropic APIs.
Official resources: It references various official Zoom GitHub repositories (e.g., github.com/zoom/*) and downloads the Zoom Cobrowse SDK from Zoom's official static resource domains.

zoom-general