The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from external sources, specifically product pages from Wix sites, to extract images and descriptions for ad generation. This represents an indirect prompt injection surface where a malicious website could provide instructions that influence the agent's behavior during prompt engineering.
Ingestion points: Browser tool output and shell curl commands directed at external product URLs as documented in references/wix-product-image-extraction.md.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are used when processing the extracted web content.
Capability inventory: The skill can write files to the local system, execute shell commands via scripts/generate.py, and interact with external APIs (Fal.ai).
Sanitization: The scripts/generate.py script applies slugification to filenames, but there is no specific sanitization for the content processed from external URLs before it is interpolated into generation prompts.
[COMMAND_EXECUTION]: The script scripts/generate.py executes external binaries, including ffmpeg and ffprobe, to process and transcode media files. These calls are implemented using list-based arguments, which is a safe practice that mitigates command injection risks. The script also requires the presence of these tools on the host system to function.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the fal-client package and performs runtime downloads of generated media files from Fal.ai CDN URLs. These operations target a well-known service and are part of the skill's core functionality for media generation.

media-gen