The Agent Skills Directory

[PROMPT_INJECTION]: The skill demonstrates processing meeting transcripts and chat messages—which are untrusted inputs from meeting participants—and passing them to language models for summarization or analysis. Ingestion points include transcripts and chat data entering through onTranscriptData and onChatData callbacks in SKILL.md and examples/ai-integration.md. The capability inventory shows the ability to transmit this data to external AI service providers. The provided code snippets do not include explicit delimiters or instructions to distinguish participant-generated content from system instructions, and no sanitization of the transcript text is implemented in the provided reference patterns.- [DATA_EXFILTRATION]: The skill handles live media streams and requires sensitive credentials for operation. The documentation correctly instructs on the use of environment variables for managing OAuth secrets and API keys, which is a secure practice. The examples involve receiving and processing binary media data (audio and video) and transmitting data to legitimate, well-known technology service endpoints.- [EXTERNAL_DOWNLOADS]: The skill references several software packages and SDKs required for implementation, including the official @zoom/rtms SDK and various well-known AI service libraries. These resources are fetched from official package registries and trusted repository sources, representing standard and expected development dependencies.

zoom-rtms