The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/fetch_icons.py script downloads PNG icons from JSDelivr (cdn.jsdelivr.net), a well-known and trusted CDN for serving open-source assets. These icons are used as static graphic elements for the video overlays.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from a config.json file which defines overlay text and layout parameters. This data is rendered into images and used in FFmpeg filter strings, representing an attack surface for influencing video output.
Ingestion points: config.json (specifically the overlays and layout keys) processed in scripts/build_overlays.py and scripts/compose_master.py.
Boundary markers: Absent; text content is rendered directly into overlay images without delimiters.
Capability inventory: File system access (read/write), subprocess execution of FFmpeg and FFprobe.
Sanitization: scripts/build_overlays.py includes validation to ensure that the overlay copy provided in the configuration is not empty.
[COMMAND_EXECUTION]: The skill uses subprocess.run to execute FFmpeg and FFprobe for video compositing, audio analysis, and metadata verification. While the commands themselves are executed using list-based arguments (a safe practice), the complex filter strings are constructed using configuration values via string interpolation.

render-proof-points-overlay