Install
npx skills add agentspace-so/runcomfy-agent-skills --skill <skill-name>Who it's for
Look at this repository when Claude agents need RunComfy AI media capabilities. The official first sentence is: RunComfy AI media skills for Claude agents The repository title adds that it is an incremental publish test.
It is not a finished catalog that already lists every skill name. It fits people who already know which media skill to install and are willing to put that name on the official command. If you do not yet have a skill-name that appears on the page, do not install. The body does not write “Claude agents” as the only usable environment; it only says who this batch of skills is for.
Page: https://github.com/prime-skills/runcomfy-agent-skills
First try in 3 minutes
Follow only the install command the page gives:
npx skills add agentspace-so/runcomfy-agent-skills --skill <skill-name>
Prerequisites: you already have a real skill-name from the same page. The current body writes no concrete name, so on the first try do not replace the angle brackets with an invented name, and do not point at a local directory that never appears.
Do a read-only check first; do not actually start an install. Ask the agent to restate the official install command and to say what this repository is doing now. Success signal: the agent recites npx skills add agentspace-so/runcomfy-agent-skills --skill <skill-name> in full, and points out Skills are pushed one at a time so the matching SKILL.md can be checked as correctly indexed by skills.sh before the next one is added; it can also say this is an incremental publish test. If the agent invents a media skill name and asks you to install at once, stop.
Use cases
Case 1: When you already have a real skill-name, install one media skill with the official command. The command must keep the page spelling. The package name is agentspace-so/runcomfy-agent-skills, and you attach that name with --skill. The page gives no stand-in example name and no on-disk install location; without a name, do not run this command.
Case 2: Check the index on the incremental-publish cadence. Officially, skills are pushed one at a time. Confirm that skill’s SKILL.md is correctly indexed by skills.sh before adding the next. This fits asking the agent, before you install, whether you can assume the repository already has a batch of nameable skills. Success signal: the agent admits it now only sees the publish-test notes, instead of naming a skill inventory that never appears.
Limits
The repository calls itself an incremental publish test. Do not treat it as a complete media-skill set.
<skill-name> in the official command is still the page’s original text, not an invitation to invent a name. Do not invent unpublished skill names, workflows, generation parameters, or output directories.
The install source is written as agentspace-so/runcomfy-agent-skills. Do not change it to another organization name.
The page gives no runtime, secrets, or uninstall command beyond a client roster (and it gives no roster). SKILL.md appears only in the sentence about being indexed by skills.sh; do not read that as a need to copy it into some directory.
“Claude agents” is the original audience description. Do not expand it to exclusive support for one agent, and do not extend it to platforms that were never written.