The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documentation proactively identifies potential indirect prompt injection vectors when processing untrusted third-party content, such as media assets or web search results.
Ingestion points: Untrusted URLs or content passed to the runcomfy run command via the --input JSON parameter.
Boundary markers: Instructions advise passing prompts as JSON strings to avoid shell expansion and provide guidance on handling untrusted assets.
Capability inventory: The skill is restricted to the runcomfy CLI for API interaction and file downloads.
Sanitization: The documentation recommends explicit user consent for asset ingestion and disabling web search by default to mitigate automated exploitation.
[COMMAND_EXECUTION]: The skill uses the Bash(runcomfy *) tool to interact with the RunComfy API. The instructions are scoped to legitimate subcommands such as login, whoami, and run, which are the primary functions of the tool.
[EXTERNAL_DOWNLOADS]: The skill references installation and operation through official vendor channels.
Recommends installation of the @runcomfy/cli package from the standard npm registry.
Downloads generated assets from *.runcomfy.net and *.runcomfy.com domains, which are identified as official service endpoints.
Explicitly cautions against piping remote installation scripts directly into the shell without review.

runcomfy-cli