# Run Image and Video Models with RunComfy

Hosted media models use different input fields and job settings. The RunComfy CLI skill helps you find the right schema, submit a request, and track the result from the terminal.

## Install

```bash
npx skillstore add prime-skills/runcomfy-cli
```

## Metadata

- Status: approved
- Slug: prime-skills-runcomfy-cli
- Skillstore revision: r1
- Version status: missing
- Tree hash: 781a102c503d4a36306eb5a57bdbebae67ad94e91d1d86420deed7cec7e1141e
- Author: prime-skills
- GitHub username: prime-skills
- License: MIT
- Repository: https://github.com/prime-skills/runcomfy-agent-skills/tree/fca19ae084c2ae4667f5c8affbb99da5103d5832/runcomfy-cli
- Ref: da7ac991a2fd8fdea5893507682cf8209e83186a
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network, filesystem
- Quality score: 50
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/prime-skills-runcomfy-cli
- Manifest: https://skillstore.pages.dev/api/skills/prime-skills-runcomfy-cli/manifest

## Capabilities

- Set up browser sign-in or use an environment token for unattended sessions.
- Find endpoint input schemas in the API tab on RunComfy model pages.
- Submit image and video jobs with inline JSON or a JSON or YAML input file.
- Wait for results or return a request ID and check its status later.
- Choose a local output directory or return results without downloading files.
- Use JSON output to extract request IDs or result URLs in scripts.

## Use Cases

- Generate a concept image: Choose a text-to-image model, submit a non-sensitive concept prompt, and download the result to a project folder.
- Track a long video job: Submit a video job without waiting, keep the request ID, and check the result when it is ready.
- Plan an image generation batch: Review a list of prompts and prepare a batch that records request IDs and output locations for each job.

## Prompt Templates

### Check my RunComfy setup

```
Check whether the RunComfy CLI is installed and signed in. Report the version and sign-in status without displaying or requesting my token.
```

### Prepare a concept image request

```
Find the required inputs for the RunComfy text-to-image model I select. Show me the prompt and output directory before submitting one job.
```

### Submit a video job for later review

```
Validate the approved video request against its model schema. Submit it without waiting, return the request ID, and check that same job when I ask.
```

### Plan a bounded image batch

```
Review my prompt file and plan up to five image requests. Use JSON output to track each job. Ask before submitting, and do not retry ambiguous timeouts automatically.
```

## Limitations

- Model processing runs on RunComfy and requires network access and an authenticated account.
- The CLI must be installed separately. The skill does not include the CLI binary or model files.
- Required input fields vary by endpoint. Check the model's API tab before submitting.
- A local timeout does not establish that a remote job stopped. Check its status before submitting again.

## Best Practices

- Check the exact endpoint schema and confirm the request contents before sending data to RunComfy.
- Pass prompt text as structured data instead of inserting it directly into shell commands.
- Keep request IDs and check their status before retrying a request that timed out.

## Anti Patterns

- Do not paste API tokens into prompts or commit them to a repository.
- Do not install unreviewed packages or pipe remote installers into a shell.
- Do not submit confidential files or unrelated reference URLs without approval.

## Security Audit

- Audited at: 2026-11-19T17:00:32.958\+00:00
- Summary: Most static alerts match Markdown formatting, links, or routine RunComfy account checks. The skill documents ordinary hosted model use, but unpinned installs, shell argument handling, remote input submission, and retries after timeouts carry risks. No evidence found of deliberate credential exfiltration or an attempt to override audit instructions.

## Stats

- Views: 0
- Downloads: 1
- Favorites: 0
- Popularity score: 0
