The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions and recipes require the execution of various shell commands, including the vendor-provided gen-ai CLI and standard utilities like curl, jq, xargs, and ffmpeg to manage video generation and post-processing tasks.
[EXTERNAL_DOWNLOADS]: The skill implements a workflow to download video assets from URLs generated by the Picsart gen-ai API using curl (e.g., jq -r '.url' | xargs curl -L -o reel.mp4).
[INDIRECT_PROMPT_INJECTION]: User-provided prompts are interpolated directly into shell command flags for the gen-ai CLI, creating a surface where untrusted input could potentially influence command execution.
Ingestion points: User-supplied text prompts for generating or editing videos, as described in the CLI vs MCP and MCP / natural-language patterns sections of SKILL.md.
Boundary markers: The provided bash recipes use double quotes to wrap the prompt strings (e.g., -p "..."), which provides a basic level of delimitation but does not entirely eliminate the injection surface.
Capability inventory: The skill possesses capabilities for CLI execution, local file writing (logs in ~/.gen-ai/history/ and output media), and network downloads via curl.
Sanitization: The instructions do not specify any explicit sanitization, validation, or escaping of the user prompt content before it is used to construct shell commands.

gen-ai-video