The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to download and install packages from an unverified third-party scope.
Evidence: npm install @helios-project/core @helios-project/renderer @helios-project/cli in STEP 2.
[COMMAND_EXECUTION]: The skill executes multiple shell commands to manage dependencies, analyze media, and render video content.
Evidence:
npm install for third-party packages in STEP 2.
ffmpeg for audio analysis in STEP 5.
npx helios render ./composition.html -o output.mp4 in STEP 9.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external URLs and local files to drive code generation and execution logic.
Ingestion points: Fetches user-provided URLs for brand research and analyzes the local codebase (STEP 3).
Boundary markers: Absent. The skill does not implement delimiters or instructions to ignore embedded commands in the fetched content.
Capability inventory: Command execution (npm, npx, ffmpeg) and file writing (composition.html).
Sanitization: Absent. The brand intelligence extracted from external sites is used directly to construct a creative specification and generate HTML/JavaScript code.
[DYNAMIC_EXECUTION]: The skill generates a JavaScript-driven HTML file at runtime and subsequently executes it via a CLI renderer.
Evidence: STEP 8 involves creating composition.html with motion.dev and Helios runtime logic, which is then processed by npx helios in STEP 9.

guided-promo-video