Agent-Skills.md

Agent Skills: ComfyUI Workflow Editing

Use when editing ComfyUI workflow JSON, adding nodes, wiring connections, modifying workflows, adding ControlNet/LoRA/upscaling to a workflow, or submitting workflows to ComfyUI.

UncategorizedID: peteromallet/vibecomfy/comfy-edit

Repository

peteromallet/vibecomfy
peteromallet
8813

Install this agent skill to your local

pnpm dlx add-skill https://github.com/peteromallet/VibeComfy/tree/HEAD/.claude/skills/comfy-edit

Skill Files

Browse the full folder contents for comfy-edit.

Download Skill

Loading file tree…

.claude/skills/comfy-edit/SKILL.md

Skill Metadata

Name
comfy-edit
Description
Use when editing ComfyUI workflow JSON, adding nodes, wiring connections, modifying workflows, adding ControlNet/LoRA/upscaling to a workflow, or submitting workflows to ComfyUI.

ComfyUI Workflow Editing

Edit workflow JSON files using the VibeComfy CLI.

Repo: https://github.com/peteromallet/VibeComfy

When working in the repo, see CLAUDE.md for full command reference.

CLI Commands

All edit commands require -o output.json. Use --dry-run to preview changes.

| Command | Purpose | |---------|---------| | trace WF NODE | Show node inputs/outputs with slot names | | upstream WF NODE | Find nodes feeding into target | | downstream WF NODE | Find nodes fed by source | | wire WF SRC_ID SRC_SLOT DST_ID DST_SLOT -o OUT | Connect nodes | | copy WF NODE -o OUT | Duplicate a node | | set WF NODE key=val -o OUT | Set widget values | | delete WF NODE -o OUT | Remove nodes | | create WF TYPE -o OUT | Create new node | | batch WF SCRIPT -o OUT | Multi-step edits | | submit WF | Run workflow in ComfyUI |

Tip: Use trace first to find slot numbers, then wire to connect.

Adding Nodes Workflow

  1. Find the node: comfy_search("controlnet") (use comfy-registry skill)
  2. Get the spec: comfy_spec("ControlNetLoader") - see inputs/outputs
  3. Understand the flow: python we_vibin.py trace workflow.json <ksampler_id>
  4. Create and wire:
python we_vibin.py create workflow.json ControlNetLoader -o step1.json
python we_vibin.py trace step1.json <new_node_id>  # Find the slot numbers
python we_vibin.py wire step1.json <src_id> <src_slot> <dst_id> <dst_slot> -o step2.json

Or use batch scripts for multi-step edits (slot names work in batch mode):

create ControlNetLoader as $cn
wire $cn:control_net -> 15:control_net

Common Patterns

What nodes go together. See references/PATTERNS.md for details.

Image Generation

| Pattern | Flow | |---------|------| | txt2img | Checkpoint → CLIP (x2) → EmptyLatent → KSampler → VAEDecode → Save | | img2img | LoadImage → VAEEncode → KSampler (denoise<1) → VAEDecode → Save | | inpaint | LoadImage + Mask → VAEEncodeForInpaint → KSampler → VAEDecode |

Enhancements

| Pattern | Flow | |---------|------| | ControlNet | LoadControlNet → ControlNetApply → KSampler | | LoRA | Checkpoint → LoraLoader → KSampler | | Upscale | LoadUpscaleModel → ImageUpscaleWithModel → Save | | IPAdapter | LoadIPAdapter → IPAdapterApply → KSampler |

Video

| Pattern | Flow | |---------|------| | AnimateDiff | Checkpoint → ADE_LoadAnimateDiff → ADE_Apply → KSampler → VHS_Combine | | Wan t2v | WanModelLoader → WanTextEncode → WanSampler → WanDecode | | LTX | LTXVLoader → LTXVTextEncode → LTXVSampler → LTXVDecode | | v2v | VHS_LoadVideo → VAEEncode → KSampler → VAEDecode → VHS_Combine |

Flux

| Pattern | Flow | |---------|------| | Flux txt2img | UNETLoader + DualCLIPLoader + VAELoader → FluxGuidance → KSampler |

Batch Scripts

For multi-step edits, use batch scripts:

# Copy node, assign to variable
copy 1183 as $selector

# Set widgets
set $selector indexes="0:81"

# Wire using slot names
wire 1140:IMAGE -> $selector:image

# Delete nodes
delete 1220 1228

Run with: python we_vibin.py batch workflow.json script.txt -o out.json --dry-run