The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external video content via the video_url parameter, creating a surface for indirect prompt injection where malicious metadata or visual patterns in the source media could attempt to influence agent behavior during processing.
Ingestion points: The video_url parameter used in shell command examples within SKILL.md.
Boundary markers: The skill utilizes JSON-formatted input strings passed via the --input flag to delimit data from commands.
Capability inventory: The skill uses the runcomfy CLI which performs network requests to remote APIs and writes files to the local system.
Sanitization: The documentation explicitly acknowledges the risk of third-party content and provides guidance for agents to suspect source videos if output diverges significantly from prompt instructions.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the @runcomfy/cli package from the npm registry. This is an official tool provided by RunComfy, a well-known service in the AI media generation space, to facilitate access to their models.
[COMMAND_EXECUTION]: Shell commands are utilized via the runcomfy CLI to trigger remote video editing jobs. The execution is scoped to the runcomfy command as defined in the skill's allowed-tools configuration, limiting the risk of arbitrary command execution.

video-inpainting