The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted video and audio data for analysis and transcription. This creates a vulnerability surface where instructions in source media could influence AI prompt generation.
Ingestion points: Video and audio files processed in the analyze-video and clone-ad workflows.
Boundary markers: Mitigated by a mandatory 'Dialogue Confirmation Gate' requiring explicit human approval of extracted text before proceeding to generation.
Capability inventory: Executes subprocesses (ffmpeg, ffprobe, whisper), Node commands (npx hyperframes), and makes external API calls to external-api.arcads.ai.
Sanitization: Human-in-the-loop validation is the primary sanitization mechanism for extracted content.
[EXTERNAL_DOWNLOADS]: Recommends the installation of trusted third-party tools and libraries for core functionality.
Evidence: Recommends installing ffmpeg via Homebrew and openai-whisper via Pip.
[COMMAND_EXECUTION]: Executes shell scripts and CLI tools for media processing.
Evidence: Uses prompting/analyze-video/scripts/extract-frames.sh to wrap ffmpeg and ffprobe for frame and audio extraction; uses npx hyperframes for video captioning workflows.

arcads-external-api