mckruz/comfyui-expert
4 skills3.2k installs276 starsGitHub
Install
npx skills add https://github.com/mckruz/comfyui-expertSkills in this repo
1Comfyui ApiComfyUI API is an integration skill for agents and solo builders who run local or remote ComfyUI servers as part of a generative media stack. It documents how to verify the server with system stats, enqueue prompts, poll history until completion, and download rendered files via the view endpoint—using curl-friendly patterns and sensible timeouts. When ComfyUI is not running, it falls back to producing valid workflow JSON you can drop into the UI manually, keeping the same project folder conventions. The skill is invoked for interactive generation loops and status checks rather than for designing node graphs from scratch. It expects basic HTTP tooling on the host and optional pairing with inventory skills to discover models and nodes before execution.1.1kinstalls2Comfyui Workflow Buildercomfyui-workflow-builder is an agent skill for solo builders and creators who need ComfyUI JSON workflows that actually run—correct node class names, wired required inputs, checkpoint filenames that exist, and resolutions suited to the chosen model. The packaged evaluation spec defines six weighted comparison criteria, from valid class_types and semantic output-index rules through dangling-input traversal, resolution fit, VRAM plausibility, and known checkpoint references. Documented uplift goals contrast a 30% baseline against an 85% with-skill target, with acceptable token overhead and fewer manual graph fixes. Use it while integrating generative media into a product demo, content pipeline, or internal tooling during Build, not as a substitute for installing ComfyUI nodes or managing GPU hardware. Agents following the skill should emit JSON graphs agents and humans can load without silent broken edges.987installs3Comfyui Video PipelineComfyUI Video Pipeline is an agent skill for solo creators and indie builders who want structured video generation without manually guessing which diffusion stack fits their GPU and creative goal. The skill acts as an orchestration layer: it walks through an engine-selection tree that weighs film-level quality, long-form duration, iteration speed, camera motion control, and first-last-frame constraints, then documents pipeline prerequisites such as specific safetensors checkpoints and VAE paths for Wan 2.2 image-to-video. It is user-invocable and assumes a running ComfyUI endpoint you address via COMFYUI_URL. Use it when your agent should produce marketing clips, character motion from stills, or text-driven scenes rather than single static images. Complexity rises with MoE 14B workflows, so the skill’s branching logic is the main value for one-person teams balancing quality against VRAM limits.588installs4Comfyui Prompt EngineerComfyUI Prompt Engineer is a generative-media agent skill that crafts checkpoint-specific positive and negative prompts for ComfyUI workflows. Solo builders shipping visual products, marketing assets, or automated content pipelines install it when an agent must switch between FLUX’s natural-language style, SDXL’s tag-heavy conventions, legacy SD1.5 habits, and Wan video syntax without blending rules that degrade quality. The SKILL.md spells CFG ranges, length guidance, and structural patterns so refinements stay consistent across batch runs and hand-offs to workflow nodes. It targets intermediate users who already run ComfyUI locally or on a GPU box and want procedural prompt knowledge rather than one-off chat guesses. Use it whenever you are generating or refining prompts tied to a named checkpoint or identity method. It does not install nodes or manage GPUs—that stays in your ComfyUI stack. Check Prism Security Audits for the parent package before enabling in agents with filesystem or shell access to model weights.554installs