The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process live meeting transcripts and chat messages, which are untrusted external inputs. In the provided integration examples (e.g., examples/ai-integration.md), these data streams are directly interpolated into LLM prompts for summarization and analysis without explicit sanitization or instruction boundary markers. This creates an attack surface for indirect prompt injection, where a meeting participant could theoretically speak or type malicious instructions intended to manipulate the AI's summarization logic or downstream actions. However, as this is the primary intended purpose of the skill (analyzing participant data) and it provides standard developer guidance for a media ingestion service, the risk is categorized as low and characteristic of the domain.
Ingestion points: Meeting transcripts, chat messages, and active speaker events received via WebSockets in examples/ai-integration.md and examples/sdk-quickstart.md.
Boundary markers: No explicit delimiters or 'ignore' instructions are shown in the prompt templates used for summarization examples.
Capability inventory: The examples demonstrate the ability to send processed data to external AI APIs (OpenAI, Deepgram, AssemblyAI) and write media/transcripts to the local file system.
Sanitization: No sanitization or filtering of text-based media content is demonstrated before LLM interpolation.

zoom-rtms