The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external data from sources such as YouTube, remote URLs, and local files to generate transcripts and searchable indices. This creates a potential attack surface where instructions embedded in media content (e.g., in speech that is later transcribed) could influence agent behavior.
Ingestion points: Media content is ingested from YouTube links, arbitrary external URLs, and local filesystem paths.
Boundary markers: The instructions do not specify any delimiters or warnings to the agent to disregard instructions potentially contained within the processed media or transcripts.
Capability inventory: The skill uses the VideoDB SDK for network-based video ingestion, file system access for local uploads, and AI media generation.
Sanitization: No methods for sanitizing, filtering, or escaping the content retrieved from external media sources are described.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the videodb Python SDK (including its capture extras) and the python-dotenv library. It also references its source repository on GitHub.

videodb-skills