The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads generated audio assets (MP3) from the ElevenLabs API (api.elevenlabs.io) for sound effects and music beds. It also fetches font files from standard content delivery networks for styling the video overlays.
[COMMAND_EXECUTION]: The skill uses subprocess.run to execute ffmpeg and ffprobe for media processing, including volume detection, silence detection, duration measurement, and peak normalization of audio files. These operations are standard for video editing workflows.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data by scraping brand websites using Puppeteer to extract brand colors, logos, and screenshots. It also processes user-provided video footage for transcription. While this creates an attack surface for indirect prompt injection, the risk is managed through manual review steps.
Ingestion points: Automated scraping of third-party websites in references/design.md and Whisper transcription of raw user video files in SKILL.md.
Boundary markers: The skill lacks explicit boundary delimiters for external data, but includes specific instructions for human verification of drafts and automated QA checks on the final render.
Capability inventory: The environment includes ffmpeg, ffprobe, puppeteer, and urllib.request network capabilities, which are used to process and write the final video output.
Sanitization: The scripts/generate-captions.mjs utility includes basic HTML entity escaping for caption text to prevent injection into the video rendering template.

branded-ad-edit