The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions authorize the use of yt-dlp and FFprobe for extracting media metadata and downloading assets. These commands are localized to the primary task of media processing and are gated by a mandatory beat-planning phase.- [EXTERNAL_DOWNLOADS]: Sourcing candidates involves research on public channels, official entity pages, and archives. The skill maintains a safe posture by requiring explicit user authorization before performing any paid acquisition, signing in with credentials, or initiating downloads.- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted transcripts and media files provided by the user. While this creates an input surface, the risk is managed by a structured workflow that mandates an annotated beat plan and human approval for all sourced assets before placement.- [DATA_EXFILTRATION]: Network operations are scoped to research and media sourcing. The instructions include a specific safeguard to ask before sending anything externally, ensuring that research remains read-only unless the user provides permission.

find-broll