The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted content from meeting participants through several interfaces, including chat notifications (onChatMsgNotification), live transcription callbacks (onLiveTranscriptionMsgInfoReceived), and AI companion query responses (onReceiveQueryAnswer). This data enters the agent's context and could be used to influence downstream actions.
Ingestion points: Examples provided in examples/chat.md, examples/captions-transcription.md, and examples/ai-companion.md show where participant-controlled data is received.
Boundary markers: The code examples do not include delimiters or instructions to treat meeting data as untrusted content.
Capability inventory: The skill enables significant control over the environment, including audio manipulation via IMeetingAudioController, chat messaging via IMeetingChatController, breakout room management via IMeetingBOController, and remote camera control via IMeetingCameraHelper.
Sanitization: There is no evidence of sanitization or validation logic for the ingested meeting content in the provided documentation.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions to download the vcpkg package manager from Microsoft's official GitHub repository and the Zoom Meeting SDK from the Zoom Marketplace. These downloads are associated with well-known and trusted technology providers.
[CREDENTIALS_UNSAFE]: The skill advises users to manage sensitive authentication details, such as JWT tokens and meeting passcodes, within a local config.json file. While the documentation correctly uses placeholders for these values in examples, the recommended practice involves storing credentials in plaintext on the local filesystem for use by the SDK.

zoom-meeting-sdk-windows