메뉴
Query remote ComfyUI servers and analyze workflow files using direct API access and local parsing.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
A specialized AI agent that ruthlessly evaluates all marketing assets against world-class standards, ensuring only premium content is approved for final delivery.
A specialized framework for generating high-fidelity prompts optimized for Flux.2 (image) and LTX 2.5 (video). Focuses on materiality, optical physics, temporal dynamics, and cinematic technicality to move beyond "vibe-coded" descriptions.
Automates the transformation of narrative Markdown prompts into technical JSON payloads for Z-Image Turbo ComfyUI workflows. Handles extraction, sanitization against banned terms, and dynamic parameter injection (Resolution/Node Targeting).
Transforms narrative script data into cinematic prompt batches for FLUX/LTX.
Expert-tier prompting protocol for Grok Imagine Video Generation (xAI). Focuses on 'Hollywood sequence' narrative prose, SCENE LOCK consistency, and handheld realism.
High-precision prompting protocol for LTX 2.3 (Lightricks) video generation, optimized for ComfyUI VFX pipelines and local control. Incorporates cinematic prose, camera motion selectors, and the community-standard 3-stage sampling hack.
| name | comfyui-workflow-query-analysis |
| version | 1.0 |
| description | Query remote ComfyUI servers and analyze workflow files using direct API access and local parsing. |
A comprehensive guide for querying remote ComfyUI servers and analyzing workflow files locally.
curl http://localhost:8188/api/system_stats | python3 -m json.tool
curl http://localhost:8189/api/system_stats | python3 -m json.tool
⚠️ Must use /api/system_stats — not /system_stats (returns nothing on ComfyUI 0.18.2+).
Response includes:
win32), Python version, PyTorch versionComfyUI has two different JSON formats. Using the wrong parser for the wrong format is a common failure.
/prompt)Flat dictionary keyed by node ID string. This is what you inject into and POST.
{
"6": { "class_type": "CLIPTextEncode", "inputs": { "text": "...", "clip": ["4", 1] } },
"8": { "class_type": "EmptyLatentImage", "inputs": { "width": 1024, "height": 1024 } }
}
Nested structure with a nodes array and a definitions.subgraphs wrapper (used in newer ComfyUI versions) or a top-level nodes array (older versions). Cannot be submitted directly to /prompt — must be converted to Format 1 first.
{
"id": "workflow-id",
"definitions": {
"subgraphs": [
{ "nodes": [...], "links": [...] }
]
}
}
Rule: Workflows in ~/ComfyUI/workflows/ saved from the UI are Format 2. Workflows used for API injection must be Format 1. When in doubt, check whether the top-level keys are numeric strings — if yes, it's Format 1.
import json
from pathlib import Path
def analyze_workflow(wf_file):
with open(wf_file) as f:
d = json.loads(f.read())
# Get nodes from subgraphs (not flat 'nodes' field)
defs = d.get('definitions', {})
subgraphs = defs.get('subgraphs', [])
all_nodes = []
for sg in subgraphs:
if isinstance(sg, dict):
nodes_list = sg.get('nodes', [])
if isinstance(nodes_list, list):
all_nodes.extend(nodes_list)
node_types = [n.get('class_type', 'unknown') for n in all_nodes]
return len(all_nodes), node_types
def analyze_api_workflow(wf_file):
with open(wf_file) as f:
d = json.load(f)
for node_id, node in d.items():
print(f"Node {node_id}: {node.get('class_type')} — inputs: {list(node.get('inputs', {}).keys())}")
Some workflows may have empty subgraphs - these are typically:
Check: Verify the workflow was fully saved by opening in ComfyUI and re-exporting.
For listing and reading workflow files, use the MCP layer (list_workflows, read_workflow) — those tools read local ~/workflows/ files without needing the server up. This skill handles the analysis and direct API query layer. See comfyui-master-control for the full two-layer architecture.
/api/system_stats