Install comfy-cli's bundled agent skills (comfy, comfy-build, comfy-debug, comfy-deploy, comfy-director, comfy-relay) into any harness and apply this toolkit's zero-cost validation gate and credit audit around paid ComfyUI runs. Use when setting up comfy-cli for an agent, when the comfy skills are missing from the skill catalog, or before submitting a workflow that bills partner-node credits.
6 stars
0 votes
0 copies
2 views
Added September 20, 2026
ai-agentsnodeapifrontend
Works with
cli
api
Security analysis
A96/100
mediumInstalls packages at runtime which could introduce malicious dependencies
Installs into .claude/skills of the current project.
Are you the author of Comfy Cli Skills?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/sideshowroberto-comfy-cli-skills)
---
name: comfy-cli-skills
description: Install comfy-cli's bundled agent skills (comfy, comfy-build, comfy-debug, comfy-deploy, comfy-director, comfy-relay) into any harness and apply this toolkit's zero-cost validation gate and credit audit around paid ComfyUI runs. Use when setting up comfy-cli for an agent, when the comfy skills are missing from the skill catalog, or before submitting a workflow that bills partner-node credits.
---
# comfy-cli skills
comfy-cli (Comfy-Org, GPL-3.0) ships its own agent skills for driving ComfyUI
from an agent: `comfy` (run and edit workflows), `comfy-build`, `comfy-debug`,
`comfy-deploy`, `comfy-director` (multi-shot narrative) and `comfy-relay` (media
presentation). They are not vendored in this MIT toolkit; install them from the
CLI so every harness reads the same text.
## Install
```sh
pip install --upgrade comfy-cli
comfy --version
comfy skills install
```
`comfy skills install` needs comfy-cli 1.20 or newer; older builds expose the
same step inside the `comfy setup` wizard. Afterwards confirm the six `comfy*`
entries appear in the harness's skill catalog (restart the client if it caches).
The skills document a 1.20 command baseline: on an older CLI, read `comfy --help`
before using a command they name.
## House rules around a PAID run
Partner nodes (Kling, Seedance, MiniMax, Veo, Gemini image) bill credits the
moment a graph is accepted. Two zero-cost steps come first, every time:
1. **Validate, then print without posting.** `comfy validate --workflow api.json
--where local` must be clean, then `comfy run --workflow api.json
--print-prompt --json` logs the exact graph. This catches wrong input names
and DynamicCombo dot-path mistakes for free.
2. **Read the balance before and after.** `GET https://api.comfy.org/customers/balance`
with header `X-API-KEY: <key>`; `effective_balance_micros` / 100 = credits.
Reconcile the delta against the price badges. A larger drop than predicted
means stop and find what else billed.
Submit with `comfy run --workflow api.json --wait --json --timeout 600`. Feed the
workflow's API-format export; UI-format graphs with Get/Set virtual nodes or
subgraphs only resolve in the frontend and cannot run headless. Keep the API key
in the environment (`COMFY_API_KEY`), never in a workflow file or a chat message.
## Native 3D template nodes are browser-only for saving
`Save3DAdvanced` and `Preview3DAdvanced` (the bundled native 3D templates,
e.g. Trellis2 image-to-model) require a `viewport_state` input of type
`LOAD_3D` that only the browser supplies, and they write nothing headless
even when the job reports `completed`. Swap in `SaveGLB` instead (mesh or
`FILE_3D` in, `filename_prefix`) - it needs no viewport state and writes the
file. Verify any headless 3D-template run by checking the outputs list for
the expected file extension; `ok: true` alone is not proof anything was
written.
## UI-to-API conversion can drop widget values on native 3D nodes
`comfy run` converts a UI-format workflow client-side, but on the native 3D
nodes some combo/int widgets arrive wrong: Trellis2UpsampleStage.target_resolution
came through as the string "1536" and the width of Preview3DAdvanced and
Save3DAdvanced as "" (four shape_mismatch errors). Fix: take the API prompt
from the run's `prompt_preview` event, patch the values, and submit that API
file instead of the UI file. Validate it with `comfy validate --where local`
before the run.