Not in the top 1,000 by GitHub stars
SlavaSexton/ComfyUI-Agent-Kit
One ComfyUI skill for every AI coding agent (Claude Code, Codex, Gemini CLI, Qwen Code): drives a local ComfyUI end to end. 75 model prompt recipes, 581 templates, plus standalone Seedance, Krea and MiniMax H3 skills. Hardware-aware selection, workflow building, multi-shot video, ControlNet, HDR. By Chronicle of Brothers.
History
Review weekly changes to SKILL.md before updating. Dates mark the weeks in which we observed each version; checks apply to the recorded commits.
View weekly changes
claude-code/skills/comfyui/SKILL.md
Week of Nov 2, 2026 · +0 / −0 lines
Before Network or broad shell → After Network or broad shell
74f5b0bbd87b → 74f5b0bbd87b
Week of Oct 26, 2026 · Line counts unavailable
Before Not checked yet → After Network or broad shell
First observation → 74f5b0bbd87b
No earlier version was observed for comparison.
claude-code/skills/krea/SKILL.md
Week of Nov 2, 2026 · +0 / −0 lines
Before Instructions only → After Instructions only
74f5b0bbd87b → 74f5b0bbd87b
Week of Oct 26, 2026 · Line counts unavailable
Before Not checked yet → After Instructions only
First observation → 74f5b0bbd87b
No earlier version was observed for comparison.
claude-code/skills/minimax-h3/SKILL.md
Week of Nov 2, 2026 · +0 / −0 lines
Before Instructions only → After Instructions only
74f5b0bbd87b → 74f5b0bbd87b
Week of Oct 26, 2026 · Line counts unavailable
Before Not checked yet → After Instructions only
First observation → 74f5b0bbd87b
No earlier version was observed for comparison.
claude-code/skills/seedance/SKILL.md
Week of Nov 2, 2026 · +0 / −0 lines
Before Instructions only → After Instructions only
74f5b0bbd87b → 74f5b0bbd87b
Week of Oct 26, 2026 · Line counts unavailable
Before Not checked yet → After Instructions only
First observation → 74f5b0bbd87b
No earlier version was observed for comparison.
shared/comfyui/SKILL.md
Week of Nov 2, 2026 · +0 / −0 lines
Before Network or broad shell → After Network or broad shell
74f5b0bbd87b → 74f5b0bbd87b
Week of Oct 26, 2026 · Line counts unavailable
Before Not checked yet → After Network or broad shell
First observation → 74f5b0bbd87b
No earlier version was observed for comparison.
shared/krea/SKILL.md
Week of Nov 2, 2026 · +0 / −0 lines
Before Instructions only → After Instructions only
74f5b0bbd87b → 74f5b0bbd87b
Week of Oct 26, 2026 · Line counts unavailable
Before Not checked yet → After Instructions only
First observation → 74f5b0bbd87b
No earlier version was observed for comparison.
shared/minimax-h3/SKILL.md
Week of Nov 2, 2026 · +0 / −0 lines
Before Instructions only → After Instructions only
74f5b0bbd87b → 74f5b0bbd87b
Week of Oct 26, 2026 · Line counts unavailable
Before Not checked yet → After Instructions only
First observation → 74f5b0bbd87b
No earlier version was observed for comparison.
shared/seedance/SKILL.md
Week of Nov 2, 2026 · +0 / −0 lines
Before Instructions only → After Instructions only
74f5b0bbd87b → 74f5b0bbd87b
Week of Oct 26, 2026 · Line counts unavailable
Before Not checked yet → After Instructions only
First observation → 74f5b0bbd87b
No earlier version was observed for comparison.
View on GitHubAdd to my skills
Verified commit 74f5b0bbd87b1c4ca0cd95dea169cdcae4b9af9d
git clone --filter=blob:none --no-checkout https://github.com/SlavaSexton/ComfyUI-Agent-Kit.git .hotskills/slavasexton-comfyui-agent-kit && git -C .hotskills/slavasexton-comfyui-agent-kit checkout --detach 74f5b0bbd87b1c4ca0cd95dea169cdcae4b9af9d
# Copy each selected skill directory from .hotskills/slavasexton-comfyui-agent-kit into .claude/skills/<skill-name>/git clone --filter=blob:none --no-checkout https://github.com/SlavaSexton/ComfyUI-Agent-Kit.git .hotskills/slavasexton-comfyui-agent-kit && git -C .hotskills/slavasexton-comfyui-agent-kit checkout --detach 74f5b0bbd87b1c4ca0cd95dea169cdcae4b9af9d
# Copy each selected skill directory from .hotskills/slavasexton-comfyui-agent-kit into .agents/skills/<skill-name>/117GitHub stars
4stars in 7 days
8skills in this repository
1commit in 30 days
Heat
4.0- Star growth, last 7 daysNot available
- Star growth, last 24 hoursNot available
- Rank movement, last 7 daysNot available
- Commits, last 30 daysNot available
- Accounts asking about it through MCP, last 7 daysNot available
- Hot votes this weekNot available
Heat never changes the order by stars, and sponsorship never touches it. How heat works
Stars over time
30-day change not yet available: not enough history.
Individual skills in this repository
- comfyui
claude-code/skills/comfyui - krea
claude-code/skills/krea - minimax-h3
claude-code/skills/minimax-h3 - seedance
claude-code/skills/seedance - comfyui
shared/comfyui - krea
shared/krea - minimax-h3
shared/minimax-h3 - seedance
shared/seedance
How it works
comfyui
When your agent uses it
Drive a local ComfyUI install for image, video, and audio generation via its HTTP API. Use WHENEVER generating, rendering, or editing images/video/audio/hero assets with ComfyUI, Z-Image, Ideogram, FLUX, LTX, Wan, or when building, parameterizing, or running ComfyUI workflows. Covers the API client, workflow JSON format, model patterns, dual/multi-GPU placement, the MCP driver, in-graph Claude nodes, and VRAM coordination.
What it loads into context
Frontmatter
117 tokensAlways available to the agent
SKILL.md
12,851 tokensWhen the skill is activated
Reference documents
245,917 tokensLoaded on request
Token counts are estimates. Scripts run separately; they are not counted as reference context.
The playbook
- Files in this kit (pull the right one on demand)
- Your machine
- The four layers (what this kit installs)
- The client (no extra deps, stdlib only)
- Workflow JSON (API format)
- Template library (the SOURCE OF TRUTH, mix and match)
- Compose a NEW workflow from pieces (assemble + wire it correctly)
- Shared workflows + model shootout (pick the best model for a look)
- Staying current (new models and workflows)
- Per-model prompting (the mega-brain): READ before prompting a named model
- In-graph Claude nodes (Layer 3): pick the right one
- Build a workflow AND show it in the owner's GUI (bidirectional bridge)
- Lay the graph out cleanly (structured blocks, no overlap)
- Collapse a stage into one reusable node (Subgraphs)
- Creator-level depth: strengths, real limits, and advanced sequence work
- Where models live, and how to download one (DETECT, do not assume)
- Pick a model variant that fits THIS machine (hardware-aware, recommend before downloading)
- Using multiple GPUs (it is NOT like a layer-split LLM server)
- VRAM coordination (CRITICAL gotcha)
- NEVER restart Comfy Desktop via the MCP (CRITICAL gotcha)
- Start ComfyUI yourself when it is down (auto-start the server)
- Session protocol (ask how to start, and SAVE so the owner can find it later)
- Procedure (do this each time)
Scripts executed separately
comfy_client.pytools/check_updates.pytools/fetch_workflow.pytools/gen_quick_index.pytools/node_inventory.pyworkflow_layout.py../../../shared/tools/fetch_workflow.py
Files
../../../.claude-plugin/marketplace.jsonasset779 bytes../../.claude-plugin/plugin.jsonasset501 bytesADVANCED.mddoc36,053 bytesAGENTS.mddoc2,926 bytesBOOTSTRAP.mddoc4,208 bytesBUILDING_NODES.mddoc19,511 bytesEXAMPLE_WORKFLOWS.mddoc2,354 bytesKIJAI.mddoc46,923 bytesKNOWN_ISSUES.mddoc21,869 bytesLAYERS.mddoc5,322 bytesLTX2_TRAINING.mddoc3,991 bytesMODELS.mddoc10,073 bytesMODELS/audio.mddoc10,313 bytesMODELS/image-api.mddoc27,285 bytesMODELS/image-edit.mddoc7,960 bytesMODELS/image-open-more.mddoc20,226 bytesMODELS/image-open.mddoc24,334 bytesMODELS/niche.mddoc12,598 bytesMODELS/three-d.mddoc3,347 bytesMODELS/utility.mddoc11,238 bytesMODELS/video-api.mddoc56,557 bytesMODELS/video-open.mddoc57,528 bytesMODEL_INDEX.mddoc7,594 bytesNODES.mddoc3,152 bytesNODE_LIBRARY/_INDEX.mddoc7,513 bytesNODE_LIBRARY/_INVENTORY.mddoc12,878 bytesNODE_LIBRARY/_SCHEMA.mddoc4,971 bytesNODE_LIBRARY/advanced.mddoc44,496 bytesNODE_LIBRARY/audio.mddoc11,437 bytesNODE_LIBRARY/color-and-transform.mddoc3,315 bytesNODE_LIBRARY/conditioning-1.mddoc43,096 bytesNODE_LIBRARY/conditioning-2.mddoc39,790 bytesNODE_LIBRARY/core.mddoc9,580 bytesNODE_LIBRARY/custom-author.mddoc9,211 bytesNODE_LIBRARY/experimental.mddoc18,021 bytesNODE_LIBRARY/image-1.mddoc33,348 bytesNODE_LIBRARY/image-2.mddoc25,653 bytesNODE_LIBRARY/latent.mddoc32,179 bytesNODE_LIBRARY/loaders.mddoc18,058 bytesNODE_LIBRARY/ocio.mddoc25,984 bytesNODE_LIBRARY/radiance.mddoc15,022 bytesNODE_LIBRARY/samplers.mddoc33,819 bytesNODE_LIBRARY/smart-upscaler.mddoc16,439 bytesNODE_LIBRARY/text.mddoc17,703 bytesNODE_LIBRARY/three-d.mddoc28,417 bytesNODE_LIBRARY/training.mddoc7,042 bytesNODE_LIBRARY/utilities.mddoc23,178 bytesNODE_LIBRARY/video.mddoc16,311 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesTASKS.mddoc8,374 bytesUPDATING.mddoc2,731 bytescomfy_client.pyruns4,367 bytesmachine.mddoc2,153 bytestools/check_updates.pyruns5,888 bytestools/fetch_workflow.pyruns1,420 bytestools/gen_quick_index.pyruns4,264 bytestools/node_inventory.pyruns6,023 bytesworkflow_layout.pyruns11,989 bytes../../../shared/tools/fetch_workflow.pyruns1,420 bytes
Read SKILL.md
# ComfyUI: driving the local install
Use this whenever the task involves generating or rendering images, video, or audio with ComfyUI, or
building/running a ComfyUI workflow. Read it first, then act.
## Files in this kit (pull the right one on demand)
Only this SKILL.md auto-loads; everything else is read when relevant, so route to it instead of leaving it unread.
**Two layouts, same files:** in the INSTALLED skill everything sits flat next to this file, so `docs/TASKS.md`
below means `TASKS.md` here and `docs/NODE_LIBRARY/ocio.md` means `NODE_LIBRARY/ocio.md`; likewise a script the docs
name under `shared/tools/` in the repo sits in `tools/` next to this file once installed. In the repo those prefixes are
literal. If a path does not resolve, drop the `docs/` and look next to this file before concluding the
file is missing.
krea
When your agent uses it
Use when a task names Krea or Krea 2, when choosing between Krea's hosted API and its open weights, when wiring the Krea 2 Image or Krea 2 Style Reference API nodes, when building with FLUX.1 Krea Dev, when someone asks about Krea Realtime or realtime or streaming video generation, or when a Krea 2 job needs ControlNet, instruction editing, identity preservation or per-layer conditioning control that core ComfyUI does not provide.
What it loads into context
Frontmatter
108 tokensAlways available to the agent
SKILL.md
3,091 tokensWhen the skill is activated
Reference documents
2,999 tokensLoaded on request
Token counts are estimates. Scripts run separately; they are not counted as reference context.
The playbook
- Which Krea for which job
- The hosted graph, buildable
- FLUX.1 Krea Dev, buildable
- Krea Realtime 14B, and why ComfyUI is not the road most travelled
- The Krea 2 custom-node ecosystem worth knowing
- When it goes wrong
Files
SKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesreference.mddoc10,601 bytes
Read SKILL.md
# Krea
Krea ships two completely separate things under one brand, and picking the wrong one wastes the job.
**The fork, decide it first.**
- **Hosted API.** Nodes `Krea2ImageNode` and `Krea2StyleReferenceNode`, category `partner/image/Krea`, billed per
image, capped at **1K output**, and the only path that reads a **Krea moodboard** from the website.
- **Open weights.** Krea 2 Raw and Turbo run locally on core nodes, free, any resolution, no moodboard.
**Who owns what, so you open one file, not three.** THIS file owns the fork, the hosted graph, the FLUX.1 Krea
Dev graph and the ecosystem map. `reference.md` next to it owns full node I/O, the price matrix, weight files
with measured sizes, and the custom-node packs. **The kit's `MODELS.md` already owns the local Krea 2 graph in
depth** (Raw vs Turbo settings, the `qwen3vl_4b` encoder and its `krea2` CLIP type, the Ostris edit method, the
style-reference LoRA on core nodes, the nine official style LoRAs, the abliterated-encoder swap, licence).minimax-h3
When your agent uses it
Use when writing or debugging prompts for MiniMax H3 (Hailuo 3) video-with-audio generation, running the open weights locally in ComfyUI, choosing a quant or an acceleration LoRA for the VRAM you have, wiring reference-to-video with images, video or audio, or when a generated clip produces gibberish speech, drifts off a reference identity, garbles audio after a latent upscale, or refuses to run on a build that looks current.
What it loads into context
Frontmatter
106 tokensAlways available to the agent
SKILL.md
2,475 tokensWhen the skill is activated
Reference documents
6,127 tokensLoaded on request
Token counts are estimates. Scripts run separately; they are not counted as reference context.
The playbook
- The prompt format is not free-form prose
- Dialogue: the single most common cause of "the speech is gibberish"
- Camera is a controlled vocabulary
- Reference-to-video: label every file with a job
- Ten production jobs to write against
- Operating facts that bite
- When it goes wrong
Files
SKILL.mdcore9,873 bytesSKILL.mdcore9,873 bytesSKILL.mdcore9,873 bytesSKILL.mdcore9,873 bytesSKILL.mdcore9,873 bytesSKILL.mdcore9,873 bytesSKILL.mdcore9,873 bytesreference.mddoc21,828 bytes
Read SKILL.md
# MiniMax H3 (Hailuo 3)
H3 generates **video and synchronised stereo audio jointly**, from text, images, reference video, reference
audio, or any mix. It is the same model family whether you hit the hosted API or run the open weights, but the
two paths are wired completely differently and only one of them is free.
**Two paths, do not confuse them.**
- **Hosted API**, partner nodes `MinimaxHailuo03TextToVideoNode` / `...FirstLastFrameNode` / `...ReferenceNode`,
category `partner/video/MiniMax`. 2K output, priced per second, no local weights.
- **Local open weights**, core nodes `MiniMaxH3ImageToVideo` and `MiniMaxH3ReferenceToVideo` from
`comfy_extras/nodes_minimax_h3.py`. 768p, free, and everything below is about this path.
**Who owns what, so you open one file, not three.** THIS file owns the prompt format and the operating rules.
`reference.md` next to it owns weights, quant sizes, acceleration packs and their wiring. The MiniMax entry inseedance
When your agent uses it
Use when writing or debugging prompts for ByteDance Seedance video models (Seedance 2.5, 2.0, 2.0 Mini, 1.5 Pro, 1.0) on Dreamina, Jimeng AI, Doubao, BytePlus ModelArk or ComfyUI, when a generated video drifts off the reference face, grows unwanted subtitles or watermarks, duplicates a character, jumps at an extension seam, or when planning multi-shot or multi-reference video briefs.
What it loads into context
Frontmatter
109 tokensAlways available to the agent
SKILL.md
2,268 tokensWhen the skill is activated
Reference documents
7,281 tokensLoaded on request
Token counts are estimates. Scripts run separately; they are not counted as reference context.
The playbook
- Which model you are talking to
- The three task types, and the word that switches between them
- Reference labels
- Four symbols that carry meaning
- Shot sequencing beats prose
- Do not fill the reference budget
- When something comes out wrong
- Sources and honesty
Files
SKILL.mdcore8,469 bytesSKILL.mdcore8,469 bytesSKILL.mdcore8,469 bytesSKILL.mdcore8,469 bytesreference.mddoc23,638 bytessupplementary.mddoc4,904 bytes
Read SKILL.md
# Seedance video prompting
ByteDance's Seedance family generates video and audio jointly and takes multimodal references
(images, video clips, audio clips) that a prompt addresses **by label**. Getting the labels and the
task type right matters more than adjectives: the same sentence can be read as "reference this" or
"edit this" depending on one word.
**Load `reference.md`** for the full formulas, the failure table, worked examples and the ComfyUI
parameters. **Load `supplementary.md`** only for non-official leads about 2.5; it never overrides
`reference.md` and carries a conflict ledger. This file is the router and the rules you need before
writing a single line.
## Which model you are talking to
comfyui
When your agent uses it
Drive a local ComfyUI install for image, video, and audio generation via its HTTP API. Use WHENEVER generating, rendering, or editing images/video/audio/hero assets with ComfyUI, Z-Image, Ideogram, FLUX, LTX, Wan, or when building, parameterizing, or running ComfyUI workflows. Covers the API client, workflow JSON format, model patterns, dual/multi-GPU placement, the MCP driver, in-graph Claude nodes, and VRAM coordination.
What it loads into context
Frontmatter
117 tokensAlways available to the agent
SKILL.md
12,851 tokensWhen the skill is activated
Reference documents
68,714 tokensLoaded on request
Token counts are estimates. Scripts run separately; they are not counted as reference context.
The playbook
- Files in this kit (pull the right one on demand)
- Your machine
- The four layers (what this kit installs)
- The client (no extra deps, stdlib only)
- Workflow JSON (API format)
- Template library (the SOURCE OF TRUTH, mix and match)
- Compose a NEW workflow from pieces (assemble + wire it correctly)
- Shared workflows + model shootout (pick the best model for a look)
- Staying current (new models and workflows)
- Per-model prompting (the mega-brain): READ before prompting a named model
- In-graph Claude nodes (Layer 3): pick the right one
- Build a workflow AND show it in the owner's GUI (bidirectional bridge)
- Lay the graph out cleanly (structured blocks, no overlap)
- Collapse a stage into one reusable node (Subgraphs)
- Creator-level depth: strengths, real limits, and advanced sequence work
- Where models live, and how to download one (DETECT, do not assume)
- Pick a model variant that fits THIS machine (hardware-aware, recommend before downloading)
- Using multiple GPUs (it is NOT like a layer-split LLM server)
- VRAM coordination (CRITICAL gotcha)
- NEVER restart Comfy Desktop via the MCP (CRITICAL gotcha)
- Start ComfyUI yourself when it is down (auto-start the server)
- Session protocol (ask how to start, and SAVE so the owner can find it later)
- Procedure (do this each time)
Scripts executed separately
comfy_client.pyworkflow_layout.py../tools/fetch_workflow.py
Files
MODELS.mddoc10,073 bytesMODELS/audio.mddoc10,313 bytesMODELS/image-api.mddoc27,285 bytesMODELS/image-edit.mddoc7,960 bytesMODELS/image-open-more.mddoc20,226 bytesMODELS/image-open.mddoc24,334 bytesMODELS/niche.mddoc12,598 bytesMODELS/three-d.mddoc3,347 bytesMODELS/utility.mddoc11,246 bytesMODELS/video-api.mddoc56,557 bytesMODELS/video-open.mddoc57,528 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytesSKILL.mdcore49,157 bytescomfy_client.pyruns4,367 bytesmachine.mddoc2,153 bytesworkflow_layout.pyruns11,989 bytesworkflows/.gitkeepasset0 bytes../tools/fetch_workflow.pyruns1,420 bytes
Read SKILL.md
# ComfyUI: driving the local install
Use this whenever the task involves generating or rendering images, video, or audio with ComfyUI, or
building/running a ComfyUI workflow. Read it first, then act.
## Files in this kit (pull the right one on demand)
Only this SKILL.md auto-loads; everything else is read when relevant, so route to it instead of leaving it unread.
**Two layouts, same files:** in the INSTALLED skill everything sits flat next to this file, so `docs/TASKS.md`
below means `TASKS.md` here and `docs/NODE_LIBRARY/ocio.md` means `NODE_LIBRARY/ocio.md`; likewise a script the docs
name under `shared/tools/` in the repo sits in `tools/` next to this file once installed. In the repo those prefixes are
literal. If a path does not resolve, drop the `docs/` and look next to this file before concluding the
file is missing.
krea
When your agent uses it
Use when a task names Krea or Krea 2, when choosing between Krea's hosted API and its open weights, when wiring the Krea 2 Image or Krea 2 Style Reference API nodes, when building with FLUX.1 Krea Dev, when someone asks about Krea Realtime or realtime or streaming video generation, or when a Krea 2 job needs ControlNet, instruction editing, identity preservation or per-layer conditioning control that core ComfyUI does not provide.
What it loads into context
Frontmatter
108 tokensAlways available to the agent
SKILL.md
3,091 tokensWhen the skill is activated
Reference documents
2,999 tokensLoaded on request
Token counts are estimates. Scripts run separately; they are not counted as reference context.
The playbook
- Which Krea for which job
- The hosted graph, buildable
- FLUX.1 Krea Dev, buildable
- Krea Realtime 14B, and why ComfyUI is not the road most travelled
- The Krea 2 custom-node ecosystem worth knowing
- When it goes wrong
Files
SKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesSKILL.mdcore11,347 bytesreference.mddoc10,601 bytes
Read SKILL.md
# Krea
Krea ships two completely separate things under one brand, and picking the wrong one wastes the job.
**The fork, decide it first.**
- **Hosted API.** Nodes `Krea2ImageNode` and `Krea2StyleReferenceNode`, category `partner/image/Krea`, billed per
image, capped at **1K output**, and the only path that reads a **Krea moodboard** from the website.
- **Open weights.** Krea 2 Raw and Turbo run locally on core nodes, free, any resolution, no moodboard.
**Who owns what, so you open one file, not three.** THIS file owns the fork, the hosted graph, the FLUX.1 Krea
Dev graph and the ecosystem map. `reference.md` next to it owns full node I/O, the price matrix, weight files
with measured sizes, and the custom-node packs. **The kit's `MODELS.md` already owns the local Krea 2 graph in
depth** (Raw vs Turbo settings, the `qwen3vl_4b` encoder and its `krea2` CLIP type, the Ostris edit method, the
style-reference LoRA on core nodes, the nine official style LoRAs, the abliterated-encoder swap, licence).minimax-h3
When your agent uses it
Use when writing or debugging prompts for MiniMax H3 (Hailuo 3) video-with-audio generation, running the open weights locally in ComfyUI, choosing a quant or an acceleration LoRA for the VRAM you have, wiring reference-to-video with images, video or audio, or when a generated clip produces gibberish speech, drifts off a reference identity, garbles audio after a latent upscale, or refuses to run on a build that looks current.
What it loads into context
Frontmatter
106 tokensAlways available to the agent
SKILL.md
2,475 tokensWhen the skill is activated
Reference documents
6,127 tokensLoaded on request
Token counts are estimates. Scripts run separately; they are not counted as reference context.
The playbook
- The prompt format is not free-form prose
- Dialogue: the single most common cause of "the speech is gibberish"
- Camera is a controlled vocabulary
- Reference-to-video: label every file with a job
- Ten production jobs to write against
- Operating facts that bite
- When it goes wrong
Files
SKILL.mdcore9,873 bytesSKILL.mdcore9,873 bytesSKILL.mdcore9,873 bytesSKILL.mdcore9,873 bytesSKILL.mdcore9,873 bytesSKILL.mdcore9,873 bytesSKILL.mdcore9,873 bytesreference.mddoc21,828 bytes
Read SKILL.md
# MiniMax H3 (Hailuo 3)
H3 generates **video and synchronised stereo audio jointly**, from text, images, reference video, reference
audio, or any mix. It is the same model family whether you hit the hosted API or run the open weights, but the
two paths are wired completely differently and only one of them is free.
**Two paths, do not confuse them.**
- **Hosted API**, partner nodes `MinimaxHailuo03TextToVideoNode` / `...FirstLastFrameNode` / `...ReferenceNode`,
category `partner/video/MiniMax`. 2K output, priced per second, no local weights.
- **Local open weights**, core nodes `MiniMaxH3ImageToVideo` and `MiniMaxH3ReferenceToVideo` from
`comfy_extras/nodes_minimax_h3.py`. 768p, free, and everything below is about this path.
**Who owns what, so you open one file, not three.** THIS file owns the prompt format and the operating rules.
`reference.md` next to it owns weights, quant sizes, acceleration packs and their wiring. The MiniMax entry inseedance
When your agent uses it
Use when writing or debugging prompts for ByteDance Seedance video models (Seedance 2.5, 2.0, 2.0 Mini, 1.5 Pro, 1.0) on Dreamina, Jimeng AI, Doubao, BytePlus ModelArk or ComfyUI, when a generated video drifts off the reference face, grows unwanted subtitles or watermarks, duplicates a character, jumps at an extension seam, or when planning multi-shot or multi-reference video briefs.
What it loads into context
Frontmatter
109 tokensAlways available to the agent
SKILL.md
2,268 tokensWhen the skill is activated
Reference documents
7,281 tokensLoaded on request
Token counts are estimates. Scripts run separately; they are not counted as reference context.
The playbook
- Which model you are talking to
- The three task types, and the word that switches between them
- Reference labels
- Four symbols that carry meaning
- Shot sequencing beats prose
- Do not fill the reference budget
- When something comes out wrong
- Sources and honesty
Files
SKILL.mdcore8,469 bytesSKILL.mdcore8,469 bytesSKILL.mdcore8,469 bytesSKILL.mdcore8,469 bytesreference.mddoc23,638 bytessupplementary.mddoc4,904 bytes
Read SKILL.md
# Seedance video prompting
ByteDance's Seedance family generates video and audio jointly and takes multimodal references
(images, video clips, audio clips) that a prompt addresses **by label**. Getting the labels and the
task type right matters more than adjectives: the same sentence can be read as "reference this" or
"edit this" depending on one word.
**Load `reference.md`** for the full formulas, the failure table, worked examples and the ComfyUI
parameters. **Load `supplementary.md`** only for non-official leads about 2.5; it never overrides
`reference.md` and carries a conflict ledger. This file is the router and the rules you need before
writing a single line.
## Which model you are talking to
Skills included
- comfyui
- krea
- minimax-h3
- seedance
- comfyui
- krea
- minimax-h3
- seedance
Repository facts
- License
- Apache-2.0
- Created
- Jun 20, 2026
- Last pushed
- Nov 2, 2026
- Forks
- 18
- Open issues
- 2
- Project activity
- Active
Before you install
Network or broad shellAutomated signals, not a guarantee
Checked Nov 6, 2026
Not fully analyzed: some files were too long, unreadable or missing. This never lowers the level.
Show 122 notes
claude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzedclaude-code/skills/comfyui/SKILL.md:1Analysis incomplete: Referenced executable target unavailable; not analyzed
- No broad tool permissions
- 1 dangerous shell commandin 1 file: comfyui/NODE_LIBRARY/radiance.md:107
- 153 network referencesin 21 files: comfyui/NODE_LIBRARY/image-2.md:7 shared/tools/fetch_workflow.py:4 comfyui/SKILL.md:170 and 18 more
- No hidden or obfuscated text
- No instruction overrides
- 10 executable scriptsin 9 files: shared/tools/fetch_workflow.py:1 comfyui/comfy_client.py:1 comfyui/tools/check_updates.py:1 and 6 more
- 18 references to secretsin 12 files: comfyui/tools/node_inventory.py:20 comfyui/NODES.md:11 comfyui/SKILL.md:252 and 9 more
- No requests to vote in Hot
Show all 182 signals by file
.claude-plugin/marketplace.json
- Network accessline 4
"owner": { "name": "AI VFX NEWS", "url": "https://aivfxnews.com/" - Network accessline 5
"name": "AI VFX NEWS", "url": "https://aivfxnews.com/" }, - Network accessline 6
"url": "https://aivfxnews.com/" - Network accessline 11
"source": "./claude-code", "description": "Drive a LOCAL ComfyUI (your GPU, no cloud, no account) end to end: 75 model prompt recipes, hardware-aware model selection, workflow building, multi-shot video, the comfyui-mcp driver, plus a - Network accessline 12
"description": "Drive a LOCAL ComfyUI (your GPU, no cloud, no account) end to end: 75 model prompt recipes, hardware-aware model selection, workflow building, multi-shot video, the comfyui-mcp driver, plus a standalone Seedance skill for By - Network accessline 13
"homepage": "https://github.com/SlavaSexton/ComfyUI-Agent-Kit"
claude-code/.claude-plugin/plugin.json
- Network accessline 5
"author": { "name": "AI VFX NEWS", "url": "https://aivfxnews.com/" - Network accessline 6
"name": "AI VFX NEWS", "url": "https://aivfxnews.com/" }, - Network accessline 7
"url": "https://aivfxnews.com/" - Network accessline 8
}, "homepage": "https://github.com/SlavaSexton/ComfyUI-Agent-Kit" } - Network accessline 9
"homepage": "https://github.com/SlavaSexton/ComfyUI-Agent-Kit"
claude-code/skills/comfyui/ADVANCED.md
- Network accessline 126
| ComfyUI-RMBG (hub) | github.com/1038lab/ComfyUI-RMBG | bundles RMBG2/BEN2/BiRefNet/SDMatte/SAM2-3 | GPL-3.0 | | LayerStyle (edge refine) | github.com/chflame163/ComfyUI_LayerStyle | MaskEdgeUltraDetailV2 ViTMatte refine | MIT | | RMBG-2.0 - Network accessline 127
| LayerStyle (edge refine) | github.com/chflame163/ComfyUI_LayerStyle | MaskEdgeUltraDetailV2 ViTMatte refine | MIT | | RMBG-2.0 | huggingface.co/briaai/RMBG-2.0 | soft-alpha segmenter | CC BY-NC (NONCOMMERCIAL) | | MatAnyone / MatAnyone2 | - Network accessline 128
| RMBG-2.0 | huggingface.co/briaai/RMBG-2.0 | soft-alpha segmenter | CC BY-NC (NONCOMMERCIAL) |
claude-code/skills/comfyui/AGENTS.md
- Network accessline 29
{ "env": { "ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic", - Network accessline 30
[Credential-related source omitted] - Network accessline 31
"ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic", - Access to secretsline 32
[Credential-related source omitted] - Network accessline 38
`claude` adapter already covers GLM**: install it for `claude` and run GLM through Claude Code; nothing GLM-specific to build. (If instead you run GLM via Cline / OpenCode / Roo with z.ai's OpenAI-compatible endpoint `https://api.z.ai/api/c - Network accessline 39
to build. (If instead you run GLM via Cline / OpenCode / Roo with z.ai's OpenAI-compatible endpoint `https://api.z.ai/api/coding/paas/v4`, register `comfyui-mcp` in that tool's own MCP settings; the knowledge in `shared/comfyui/SKILL.md` st - Network accessline 40
`https://api.z.ai/api/coding/paas/v4`, register `comfyui-mcp` in that tool's own MCP settings; the knowledge in
claude-code/skills/comfyui/BOOTSTRAP.md
- Access to secretsline 41
[Credential-related source omitted]
claude-code/skills/comfyui/EXAMPLE_WORKFLOWS.md
claude-code/skills/comfyui/LTX2_TRAINING.md
claude-code/skills/comfyui/MODELS/video-open.md
- Network accessline 518
cross-shot consistency a prompt or the Ingredients IC-LoRA does not give, mention they can train a LoRA and offer the setup (one offer, not a nag). Full guide + the offer trigger: `docs/LTX2_TRAINING.md`. - **Source:** https://ltx.io/blog - Network accessline 520
- **Source:** https://ltx.io/blog/ltx-2-3-prompt-guide (official prompt guide) ; docs.comfy.org/tutorials/video/ltx/ltx-2-3 ; huggingface.co/Lightricks/LTX-2.3 ; github.com/Lightricks/ComfyUI-LTXVideo.
claude-code/skills/comfyui/NODES.md
claude-code/skills/comfyui/NODE_LIBRARY/custom-author.md
- Access to secretsline 42
[Credential-related source omitted]
claude-code/skills/comfyui/NODE_LIBRARY/image-2.md
- Network accessline 7
### LoadImage (display: "Load Image") - **pack / source:** core ComfyUI (`nodes`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 8
- **pack / source:** core ComfyUI (`nodes`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 22
### PreviewImage (display: "Preview Image") - **pack / source:** core ComfyUI (`nodes`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 23
- **pack / source:** core ComfyUI (`nodes`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 38
### SaveAnimatedWEBP (display: "Save Animated WEBP") - **pack / source:** core ComfyUI (`comfy_extras.nodes_images`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 39
- **pack / source:** core ComfyUI (`comfy_extras.nodes_images`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 56
### SaveSVGNode (display: "Save SVG") - **pack / source:** core ComfyUI (`comfy_extras.nodes_images`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 57
- **pack / source:** core ComfyUI (`comfy_extras.nodes_images`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 70
### ResizeImageMaskNode (display: "Resize Image/Mask") - **pack / source:** core ComfyUI (`comfy_extras.nodes_post_processing`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 71
- **pack / source:** core ComfyUI (`comfy_extras.nodes_post_processing`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 86
### ResizeAndPadImage (display: "Resize And Pad Image") - **pack / source:** core ComfyUI (`comfy_extras.nodes_images`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 87
- **pack / source:** core ComfyUI (`comfy_extras.nodes_images`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 104
### RepeatImageBatch (display: "Repeat Image Batch") - **pack / source:** core ComfyUI (`comfy_extras.nodes_images`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 105
- **pack / source:** core ComfyUI (`comfy_extras.nodes_images`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 119
### SolidMask (display: "Create Solid Mask") - **pack / source:** core ComfyUI (`comfy_extras.nodes_mask`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 120
- **pack / source:** core ComfyUI (`comfy_extras.nodes_mask`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 135
### MaskComposite (display: "Combine Masks") - **pack / source:** core ComfyUI (`comfy_extras.nodes_mask`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 136
- **pack / source:** core ComfyUI (`comfy_extras.nodes_mask`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 153
### ThresholdMask (display: "Threshold Mask") - **pack / source:** core ComfyUI (`comfy_extras.nodes_mask`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 154
- **pack / source:** core ComfyUI (`comfy_extras.nodes_mask`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 168
### MaskToImage (display: "Convert Mask to Image") - **pack / source:** core ComfyUI (`comfy_extras.nodes_mask`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 169
- **pack / source:** core ComfyUI (`comfy_extras.nodes_mask`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 182
### JoinImageWithAlpha (display: "Join Image with Alpha") - **pack / source:** core ComfyUI (`comfy_extras.nodes_compositing`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 183
- **pack / source:** core ComfyUI (`comfy_extras.nodes_compositing`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 197
### MaskPreview (display: "Preview Mask") - **pack / source:** core ComfyUI (`comfy_extras.nodes_mask`) | https://github.com/comfyanonymous/ComfyUI - Network accessline 198
- **pack / source:** core ComfyUI (`comfy_extras.nodes_mask`) | https://github.com/comfyanonymous/ComfyUI
claude-code/skills/comfyui/NODE_LIBRARY/ocio.md
- Network accessline 3
**Eleven** OpenColorIO nodes we build and maintain, modelled 1:1 on The Foundry Nuke's node set. Pack: **`ComfyUI-OCIO`**, author **Slava Sexton** (https://github.com/SlavaSexton/ComfyUI-OCIO), MIT, **released v1.3.0 - Network accessline 4
**`ComfyUI-OCIO`**, author **Slava Sexton** (https://github.com/SlavaSexton/ComfyUI-OCIO), MIT, **released v1.3.0 - Network accessline 119
**Install:** `git clone https://github.com/SlavaSexton/ComfyUI-OCIO` into `custom_nodes`, then
claude-code/skills/comfyui/NODE_LIBRARY/radiance.md
- Dangerous shell operationline 107
**`@PromptServer.instance.routes.post('/radiance/terminal')`** that `exec()`s into a persistent namespace on a
claude-code/skills/comfyui/SKILL.md
- Network accessline 170
Beyond the named template library, ComfyHub hosts thousands of community-shared workflows at `comfy.org/workflows/<hash>`. Any ComfyHub share downloads as plain JSON from a predictable URL: `https://comfy.org/workflows/download/<hash>.json` - Network accessline 171
`comfy.org/workflows/<hash>`. Any ComfyHub share downloads as plain JSON from a predictable URL: `https://comfy.org/workflows/download/<hash>.json`. So you can grab any shared workflow on demand, then read or run it. Helper: `python shared/ - Network accessline 172
`https://comfy.org/workflows/download/<hash>.json`. So you can grab any shared workflow on demand, then read or run - Network accessline 193
ComfyUI ships new models constantly, and they land in the template library first. To see what is new: `git pull` the templates clone and regenerate the quick index (`gen_quick_index.py`), then DIFF the model list (names not seen before = ne - Network accessline 194
the templates clone and regenerate the quick index (`gen_quick_index.py`), then DIFF the model list (names not seen before = new models / new templates). Also read the announcements RSS at `https://blog.comfy.org/feed`. The kit ships `share - Network accessline 195
before = new models / new templates). Also read the announcements RSS at `https://blog.comfy.org/feed`. The kit - Access to secretsline 252
[Credential-related source omitted] - Access to secretsline 256
[Credential-related source omitted] - Network accessline 342
**How to download (Desktop-safe):** - Direct download is most reliable: `curl -fL -C - -o "<root>/<type>/<filename>" "<url>"`. Use the official - Network accessline 343
- Direct download is most reliable: `curl -fL -C - -o "<root>/<type>/<filename>" "<url>"`. Use the official - Network accessline 461
- **Showing the owner the running server:** the headless server already serves the full ComfyUI web UI at `http://127.0.0.1:8188`. To let the owner SEE the canvas or what you built, tell them to open that URL in a - Network accessline 462
`http://127.0.0.1:8188`. To let the owner SEE the canvas or what you built, tell them to open that URL in a - Network accessline 474
**Starting (ask first when ComfyUI is down).** If :8188 is already up, just use it (the owner has ComfyUI open, or a server runs) and do NOT start another. If it is down, ASK once: "open ComfyUI yourself and I connect, or should I start the - Network accessline 475
a server runs) and do NOT start another. If it is down, ASK once: "open ComfyUI yourself and I connect, or should I start the server headless (you peek at `http://127.0.0.1:8188` in a browser)?" Follow their choice; if they say "just auto-s - Network accessline 476
start the server headless (you peek at `http://127.0.0.1:8188` in a browser)?" Follow their choice; if they say
claude-code/skills/comfyui/UPDATING.md
- Network accessline 10
templates instantly. - **blog.comfy.org/feed**: RSS of announcements (day-0 model support, "wrapped" recaps, releases). Verified working (`https://blog.comfy.org/feed`). - Network accessline 11
- **blog.comfy.org/feed**: RSS of announcements (day-0 model support, "wrapped" recaps, releases). Verified working (`https://blog.comfy.org/feed`). - *(optional)* GitHub release feeds for ComfyUI core and node packs: `<repo>/releases.ato - Network accessline 12
working (`https://blog.comfy.org/feed`).
claude-code/skills/comfyui/comfy_client.py
- Executable scriptline 1
#!/usr/bin/env python3 - Network accessline 7
ComfyUI Desktop serves the API at http://127.0.0.1:8188 by default. - Access to secretsline 22
[Credential-related source omitted] - Network accessline 25
def _get(path): with urllib.request.urlopen(f"http://{HOST}{path}", timeout=15) as r: - Network accessline 26
with urllib.request.urlopen(f"http://{HOST}{path}", timeout=15) as r: - Network accessline 30
def _post(path, payload): req = urllib.request.Request( - Network accessline 31
req = urllib.request.Request( - Network accessline 32
f"http://{HOST}{path}", - Network accessline 34
headers={"Content-Type": "application/json"}, ) with urllib.request.urlopen(req, timeout=30) as r: - Network accessline 35
) with urllib.request.urlopen(req, timeout=30) as r: return json.load(r) - Network accessline 36
with urllib.request.urlopen(req, timeout=30) as r: - Network accessline 89
"type": item.get("type", "output"), }) with urllib.request.urlopen(f"http://{HOST}/view?{q}", timeout=120) as r: - Network accessline 90
}) with urllib.request.urlopen(f"http://{HOST}/view?{q}", timeout=120) as r: data = r.read() - Network accessline 91
with urllib.request.urlopen(f"http://{HOST}/view?{q}", timeout=120) as r:
claude-code/skills/comfyui/machine.md
- Network accessline 12
- **ComfyUI**: `<Desktop or source install>`, core path `<detect>`, API at **`http://127.0.0.1:8188`**
claude-code/skills/comfyui/tools/check_updates.py
- Executable scriptline 1
#!/usr/bin/env python3 - Network accessline 30
BLOG_FEED = "https://blog.comfy.org/feed" - Network accessline 56
def _blog(n=8): try: req = urllib.request.Request(BLOG_FEED, headers={"User-Agent": "comfyui-agent-kit"}) - Network accessline 57
try: req = urllib.request.Request(BLOG_FEED, headers={"User-Agent": "comfyui-agent-kit"}) with urllib.request.urlopen(req, timeout=30) as r: - Network accessline 58
req = urllib.request.Request(BLOG_FEED, headers={"User-Agent": "comfyui-agent-kit"}) - Network accessline 59
with urllib.request.urlopen(req, timeout=30) as r: - Access to secretsline 145
[Credential-related source omitted]
claude-code/skills/comfyui/tools/fetch_workflow.py
- Executable scriptline 1
#!/usr/bin/env python3 - Network accessline 4
ComfyHub serves every shared workflow's graph at a predictable URL derived from its hash: https://comfy.org/workflows/download/<hash>.json - Network accessline 5
https://comfy.org/workflows/download/<hash>.json - Network accessline 19
BASE = "https://comfy.org/workflows/download/{h}.json?filename={h}" - Network accessline 22
def fetch(h, outdir="."): - Network accessline 23
h = h.strip().split("-")[0] # accept "<hash>-<hash>" form url = BASE.format(h=h) req = urllib.request.Request(url, headers={"User-Agent": "comfyui-agent-kit"}) - Network accessline 24
url = BASE.format(h=h) req = urllib.request.Request(url, headers={"User-Agent": "comfyui-agent-kit"}) with urllib.request.urlopen(req, timeout=60) as r: - Network accessline 25
req = urllib.request.Request(url, headers={"User-Agent": "comfyui-agent-kit"}) - Network accessline 26
with urllib.request.urlopen(req, timeout=60) as r: - Network accessline 37
print(__doc__) raise SystemExit(2) out = fetch(sys.argv[1], sys.argv[2] if len(sys.argv) > 2 else ".") - Network accessline 38
raise SystemExit(2) out = fetch(sys.argv[1], sys.argv[2] if len(sys.argv) > 2 else ".") print("saved:", out, f"({os.path.getsize(out)} bytes)") - Network accessline 39
out = fetch(sys.argv[1], sys.argv[2] if len(sys.argv) > 2 else ".")
claude-code/skills/comfyui/tools/gen_quick_index.py
claude-code/skills/comfyui/tools/node_inventory.py
- Executable scriptline 1
#!/usr/bin/env python3 - Network accessline 13
Config via env (sensible defaults below): COMFY_URL default http://127.0.0.1:8188 - Network accessline 14
COMFY_URL default http://127.0.0.1:8188 - Network accessline 20
[Credential-related source omitted] - Access to secretsline 20
[Credential-related source omitted] - Access to secretsline 21
[Credential-related source omitted] - Access to secretsline 24
[Credential-related source omitted] - Network accessline 58
oi = json.load(urllib.request.urlopen(COMFY_URL + "/object_info", timeout=60)) - Access to secretsline 64
[Credential-related source omitted]
claude-code/skills/comfyui/workflow_layout.py
- Executable scriptline 1
#!/usr/bin/env python3
shared/comfyui/MODELS/video-open.md
- Network accessline 518
cross-shot consistency a prompt or the Ingredients IC-LoRA does not give, mention they can train a LoRA and offer the setup (one offer, not a nag). Full guide + the offer trigger: `docs/LTX2_TRAINING.md`. - **Source:** https://ltx.io/blog - Network accessline 520
- **Source:** https://ltx.io/blog/ltx-2-3-prompt-guide (official prompt guide) ; docs.comfy.org/tutorials/video/ltx/ltx-2-3 ; huggingface.co/Lightricks/LTX-2.3 ; github.com/Lightricks/ComfyUI-LTXVideo.
shared/comfyui/SKILL.md
- Network accessline 170
Beyond the named template library, ComfyHub hosts thousands of community-shared workflows at `comfy.org/workflows/<hash>`. Any ComfyHub share downloads as plain JSON from a predictable URL: `https://comfy.org/workflows/download/<hash>.json` - Network accessline 171
`comfy.org/workflows/<hash>`. Any ComfyHub share downloads as plain JSON from a predictable URL: `https://comfy.org/workflows/download/<hash>.json`. So you can grab any shared workflow on demand, then read or run it. Helper: `python shared/ - Network accessline 172
`https://comfy.org/workflows/download/<hash>.json`. So you can grab any shared workflow on demand, then read or run - Network accessline 193
ComfyUI ships new models constantly, and they land in the template library first. To see what is new: `git pull` the templates clone and regenerate the quick index (`gen_quick_index.py`), then DIFF the model list (names not seen before = ne - Network accessline 194
the templates clone and regenerate the quick index (`gen_quick_index.py`), then DIFF the model list (names not seen before = new models / new templates). Also read the announcements RSS at `https://blog.comfy.org/feed`. The kit ships `share - Network accessline 195
before = new models / new templates). Also read the announcements RSS at `https://blog.comfy.org/feed`. The kit - Access to secretsline 252
[Credential-related source omitted] - Access to secretsline 256
[Credential-related source omitted] - Network accessline 342
**How to download (Desktop-safe):** - Direct download is most reliable: `curl -fL -C - -o "<root>/<type>/<filename>" "<url>"`. Use the official - Network accessline 343
- Direct download is most reliable: `curl -fL -C - -o "<root>/<type>/<filename>" "<url>"`. Use the official - Network accessline 461
- **Showing the owner the running server:** the headless server already serves the full ComfyUI web UI at `http://127.0.0.1:8188`. To let the owner SEE the canvas or what you built, tell them to open that URL in a - Network accessline 462
`http://127.0.0.1:8188`. To let the owner SEE the canvas or what you built, tell them to open that URL in a - Network accessline 474
**Starting (ask first when ComfyUI is down).** If :8188 is already up, just use it (the owner has ComfyUI open, or a server runs) and do NOT start another. If it is down, ASK once: "open ComfyUI yourself and I connect, or should I start the - Network accessline 475
a server runs) and do NOT start another. If it is down, ASK once: "open ComfyUI yourself and I connect, or should I start the server headless (you peek at `http://127.0.0.1:8188` in a browser)?" Follow their choice; if they say "just auto-s - Network accessline 476
start the server headless (you peek at `http://127.0.0.1:8188` in a browser)?" Follow their choice; if they say
shared/comfyui/comfy_client.py
- Executable scriptline 1
#!/usr/bin/env python3 - Network accessline 7
ComfyUI Desktop serves the API at http://127.0.0.1:8188 by default. - Access to secretsline 22
[Credential-related source omitted] - Network accessline 25
def _get(path): with urllib.request.urlopen(f"http://{HOST}{path}", timeout=15) as r: - Network accessline 26
with urllib.request.urlopen(f"http://{HOST}{path}", timeout=15) as r: - Network accessline 30
def _post(path, payload): req = urllib.request.Request( - Network accessline 31
req = urllib.request.Request( - Network accessline 32
f"http://{HOST}{path}", - Network accessline 34
headers={"Content-Type": "application/json"}, ) with urllib.request.urlopen(req, timeout=30) as r: - Network accessline 35
) with urllib.request.urlopen(req, timeout=30) as r: return json.load(r) - Network accessline 36
with urllib.request.urlopen(req, timeout=30) as r: - Network accessline 89
"type": item.get("type", "output"), }) with urllib.request.urlopen(f"http://{HOST}/view?{q}", timeout=120) as r: - Network accessline 90
}) with urllib.request.urlopen(f"http://{HOST}/view?{q}", timeout=120) as r: data = r.read() - Network accessline 91
with urllib.request.urlopen(f"http://{HOST}/view?{q}", timeout=120) as r:
shared/comfyui/machine.md
- Network accessline 12
- **ComfyUI**: `<Desktop or source install>`, core path `<detect>`, API at **`http://127.0.0.1:8188`**
shared/comfyui/workflow_layout.py
- Executable scriptline 1
#!/usr/bin/env python3
shared/tools/fetch_workflow.py
- Executable scriptline 1
#!/usr/bin/env python3 - Executable scriptline 1
#!/usr/bin/env python3 - Network accessline 4
ComfyHub serves every shared workflow's graph at a predictable URL derived from its hash: https://comfy.org/workflows/download/<hash>.json - Network accessline 4
ComfyHub serves every shared workflow's graph at a predictable URL derived from its hash: https://comfy.org/workflows/download/<hash>.json - Network accessline 5
https://comfy.org/workflows/download/<hash>.json - Network accessline 5
https://comfy.org/workflows/download/<hash>.json - Network accessline 19
BASE = "https://comfy.org/workflows/download/{h}.json?filename={h}" - Network accessline 19
BASE = "https://comfy.org/workflows/download/{h}.json?filename={h}" - Network accessline 22
def fetch(h, outdir="."): - Network accessline 22
def fetch(h, outdir="."): - Network accessline 23
h = h.strip().split("-")[0] # accept "<hash>-<hash>" form url = BASE.format(h=h) req = urllib.request.Request(url, headers={"User-Agent": "comfyui-agent-kit"}) - Network accessline 23
h = h.strip().split("-")[0] # accept "<hash>-<hash>" form url = BASE.format(h=h) req = urllib.request.Request(url, headers={"User-Agent": "comfyui-agent-kit"}) - Network accessline 24
url = BASE.format(h=h) req = urllib.request.Request(url, headers={"User-Agent": "comfyui-agent-kit"}) with urllib.request.urlopen(req, timeout=60) as r: - Network accessline 24
url = BASE.format(h=h) req = urllib.request.Request(url, headers={"User-Agent": "comfyui-agent-kit"}) with urllib.request.urlopen(req, timeout=60) as r: - Network accessline 25
req = urllib.request.Request(url, headers={"User-Agent": "comfyui-agent-kit"}) - Network accessline 25
req = urllib.request.Request(url, headers={"User-Agent": "comfyui-agent-kit"}) - Network accessline 26
with urllib.request.urlopen(req, timeout=60) as r: - Network accessline 26
with urllib.request.urlopen(req, timeout=60) as r: - Network accessline 37
print(__doc__) raise SystemExit(2) out = fetch(sys.argv[1], sys.argv[2] if len(sys.argv) > 2 else ".") - Network accessline 37
print(__doc__) raise SystemExit(2) out = fetch(sys.argv[1], sys.argv[2] if len(sys.argv) > 2 else ".") - Network accessline 38
raise SystemExit(2) out = fetch(sys.argv[1], sys.argv[2] if len(sys.argv) > 2 else ".") print("saved:", out, f"({os.path.getsize(out)} bytes)") - Network accessline 38
raise SystemExit(2) out = fetch(sys.argv[1], sys.argv[2] if len(sys.argv) > 2 else ".") print("saved:", out, f"({os.path.getsize(out)} bytes)") - Network accessline 39
out = fetch(sys.argv[1], sys.argv[2] if len(sys.argv) > 2 else ".") - Network accessline 39
out = fetch(sys.argv[1], sys.argv[2] if len(sys.argv) > 2 else ".")
How the signals fit together
Facts from a static read of the files at the pinned commit. They describe what the code does, not why.
Some files could not be read; the facts below still hold.
Signal confidence: 171 facts
- Network access in prose of documentationcomfyui/ADVANCED.md:126Low confidence
- Network access in prose of documentationcomfyui/ADVANCED.md:127Low confidence
- Network access in prose of documentationcomfyui/ADVANCED.md:128Low confidence
- Network access in code of documentationcomfyui/AGENTS.md:29Low confidence
- Network access in code of documentationcomfyui/AGENTS.md:30Low confidence
- Network access in code of documentationcomfyui/AGENTS.md:31Low confidence
- Access to secrets in code of documentationcomfyui/AGENTS.md:32Low confidence
- Network access in prose of documentationcomfyui/AGENTS.md:38Low confidence
- Network access in prose of documentationcomfyui/AGENTS.md:39Low confidence
- Network access in prose of documentationcomfyui/AGENTS.md:40Low confidence
- Access to secrets in prose of documentation: a variable, placeholder or call, not a stored valuecomfyui/BOOTSTRAP.md:41Low confidence
- Network access in code of documentationcomfyui/EXAMPLE_WORKFLOWS.md:7Low confidence
- Network access in code of documentationcomfyui/EXAMPLE_WORKFLOWS.md:8Low confidence
- Access to secrets in prose of documentationcomfyui/LTX2_TRAINING.md:46Low confidence
- Network access in code of documentationcomfyui/LTX2_TRAINING.md:53Low confidence
- Network access in prose of documentationcomfyui/MODELS/video-open.md:518Low confidence
- Network access in prose of documentationcomfyui/MODELS/video-open.md:520Low confidence
- Access to secrets in prose of documentationcomfyui/NODES.md:11Low confidence
- Access to secrets in prose of documentationcomfyui/NODES.md:17Low confidence
- Access to secrets in prose of documentationcomfyui/NODE_LIBRARY/custom-author.md:42Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:7Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:8Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:22Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:23Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:38Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:39Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:56Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:57Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:70Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:71Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:86Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:87Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:104Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:105Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:119Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:120Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:135Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:136Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:153Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:154Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:168Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:169Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:182Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:183Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:197Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/image-2.md:198Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/ocio.md:3Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/ocio.md:4Low confidence
- Network access in prose of documentationcomfyui/NODE_LIBRARY/ocio.md:119Low confidence
- Dangerous shell operation in prose of documentationcomfyui/NODE_LIBRARY/radiance.md:107Low confidence
and 121 more
Works with
| Agent | Status | Evidence | Install to |
|---|---|---|---|
| Claude Code | Native | Manifest .claude-plugin/marketplace.json | .claude/skills/<skill-name>/ |
| Codex | Native | Manifest claude-code/skills/comfyui/AGENTS.md | .agents/skills/<skill-name>/ |
| Cursor | Compatible | Standard SKILL.md claude-code/skills/comfyui/SKILL.md | .cursor/skills/<skill-name>/ |
| Gemini CLI | Compatible | Standard SKILL.md claude-code/skills/comfyui/SKILL.md | .gemini/skills/<skill-name>/ |
| GitHub Copilot | Compatible | Standard SKILL.md claude-code/skills/comfyui/SKILL.md | .github/skills/<skill-name>/ |
| OpenCode | Compatible | Standard SKILL.md claude-code/skills/comfyui/SKILL.md | .opencode/skills/<skill-name>/ |
- Native The repository ships a manifest, folder or frontmatter field for this agent.
- Compatible A standard SKILL.md that this agent can read.
- Unknown No evidence either way.
Derived automatically from paths, manifests and frontmatter. Not tested on each agent.
