The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates a complex media pipeline by executing several local Python scripts including render_hf.py, grade_brolls_v4.py, and build_master_v4.py. It also relies heavily on the ffmpeg command-line tool for video encoding, concatenation, and audio normalization.
[DYNAMIC_EXECUTION]: The rendering engine uses Playwright to load HTML 'hyperframes' and execute JavaScript animations via the Web Animations API (window.renderAt(t)). This allows the agent to generate video frames deterministically from web content.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it ingests user-defined content such as headlines, body text, and callouts from the configuration file. These values are interpolated into HTML templates rendered by a headless browser. While this could theoretically be used for cross-site scripting (XSS) within the isolated Playwright environment, the impact is limited to the rendering of the specific video asset.
Ingestion points: User-supplied text in beats[].headline, beats[].body, and beats[].callout fields in the configuration.
Boundary markers: No specific boundary markers or sanitization logic is described for the HTML template interpolation.
Capability inventory: Access to the local file system (via FFmpeg/Python) and headless browser execution (via Playwright).
Sanitization: The provided instructions do not specify any escaping or validation mechanisms for the text injected into the hyperframes.

render-vo-anchored-motion-listicle