The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, including scraped content from brand websites and video transcripts, which are used to generate visual overlays and captions. While basic sanitization is performed, this remains a potential vector for indirect prompt injection.\n
Ingestion points: Brand URLs (scraped via Puppeteer) and video transcripts (processed in scripts/generate-captions.mjs).\n
Boundary markers: No specific boundary markers or instructions to ignore embedded commands were identified in the template interpolation logic.\n
Capability inventory: The skill utilizes subprocess to call FFmpeg, Puppeteer for browser automation, and HyperFrames for rendering composition.\n
Sanitization: The generate-captions.mjs script employs a basic HTML escaping function to sanitize transcript text before injecting it into the composition.\n- [EXTERNAL_DOWNLOADS]: The skill fetches audio assets from the ElevenLabs API during the sound design phase.\n
Evidence: The scripts/generate-sfx.py script uses urllib.request.urlopen to download generated sound effects from https://api.elevenlabs.io/v1/sound-generation. These files are subsequently processed by local tools.\n- [DYNAMIC_EXECUTION]: The skill generates JavaScript and HTML files at runtime to define the video composition and motion graphics.\n
Evidence: scripts/generate-captions.mjs generates JavaScript tween statements that are injected into the index-template.html. This generated content is interpreted and rendered by the HyperFrames engine.\n- [CREDENTIALS_UNSAFE]: The skill requires an API key for the ElevenLabs service.\n
Evidence: SKILL.md and scripts/generate-sfx.py instruct users to provide the ELEVENLABS_API_KEY through environment variables or a .env file at the repository root, which is a standard and safe method for credential management.\n- [COMMAND_EXECUTION]: The skill invokes local command-line tools for media processing and quality assurance.\n
Evidence: The skill uses subprocess.run to execute FFmpeg and ffprobe for video cropping, audio normalization, and metadata inspection. It also invokes npx hyperframes and npm for rendering and QA tasks.

branded-ad-edit