# Comfyui Local Runner

> Run an existing ComfyUI API-format workflow on a local server from a reproducible JSON manifest. Use after local execution is selected and the exact workflow, input files, overrides, output, and failure fallback are approved.

- Skill: `kazeizumi/comfyui-local-runner` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add kazeizumi/comfyui-local-runner`
- Raw SKILL.md: https://api.skillmd.com/api/skills/kazeizumi/comfyui-local-runner/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Integrations & APIs
- Author: kazeizumi (https://skillmd.com/u/kazeizumi)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/kazeizumi/comfyui-local-runner

---


# ComfyUI local runner

This backend submits approved work. It does not start ComfyUI, install nodes,
convert UI-format workflows, or create prompts.

1. Export the workflow with **Save (API Format)** from ComfyUI.
2. Create a manifest using [references/manifest.md](references/manifest.md).
3. Validate it without contacting ComfyUI:

```bash
python scripts/run_local.py local-job.json --dry-run
```

4. Confirm the generation contract in `h3-runtime-router`, then run:

```bash
python scripts/run_local.py local-job.json
```

The runner uploads declared media, patches exact node input fields, submits one
prompt, polls its prompt ID, downloads all output files and writes `run-state.json`.
It only accepts loopback ComfyUI URLs unless `--allow-remote` is explicit.

On a timeout, inspect the saved prompt ID and ComfyUI history before retrying.
Do not create a duplicate paid or long-running job merely because the client
lost its connection.