The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources such as YouTube and public video/audio URLs, which can lead to indirect prompt injection if the ingested content contains malicious instructions.
Ingestion points: External media content is ingested via the upload_media(url) tool and the coll.upload(url=...) SDK method, then processed into text using generate_transcript.
Boundary markers: The instructions do not define boundary markers or specific guidance for the agent to treat generated transcripts as untrusted data or to ignore potential instructions embedded within the media.
Capability inventory: The skill possesses capabilities for Python code execution, local file system writes (to the .env file), and network operations via the VideoDB API.
Sanitization: No sanitization or validation procedures are specified for the media content or the resulting transcripts before they are processed by the agent.
[COMMAND_EXECUTION]: The skill provides specific Python code templates for the agent to execute in order to interact with the VideoDB SDK for uploading media, generating transcripts, and applying subtitles.

video-subtitles