The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides implementation patterns for Zoom Chat and Command Channel features. These interfaces allow the application to receive text and data from other participants in a session, creating an ingestion surface for untrusted data. If an agent processes this data without strict sanitization, it could be vulnerable to malicious instructions embedded within the messages.
Ingestion points: IZoomVideoSDKDelegate::onChatNewMessageNotify and IZoomVideoSDKDelegate::onCommandReceived in SKILL.md and examples/command-channel.md.
Boundary markers: The provided examples do not demonstrate the use of delimiters or explicit instructions to ignore embedded commands for processed content.
Capability inventory: The skill utilizes SDK singletons for session control and media manipulation, providing a surface where data ingestion could lead to unauthorized actions.
Sanitization: No data validation or escaping logic is included in the snippets processing incoming participant strings.
[EXTERNAL_DOWNLOADS]: The documentation guides users to download and install CMake from the Kitware apt repository and reference sample code from various GitHub repositories.
Fetches repository signing keys and configurations from Kitware, a well-known software service.
References project templates from external GitHub repositories for Qt and GTK integration.
[PRIVILEGE_ESCALATION]: The skill includes instructions to use elevated privileges for installing system libraries and configuring the package manager.
Evidence: sudo apt install and sudo tee commands used in SKILL.md and troubleshooting/build-errors.md for environment preparation.
[COMMAND_EXECUTION]: The skill uses shell commands to manage system configurations and set up library symlinks required for the Zoom SDK environment.
Evidence: Use of ln -sf for Qt5 library management and PulseAudio configuration commands in SKILL.md and linux.md.

video-sdk/linux