Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Introduce ComfyClaw to a user and explain how to use it end-to-end — the ComfyUI panel, run modes, CLI, skills, verifier, and the troubleshooting paths for a red 🔴 status badge. Load this skill when the user asks "what is ComfyClaw?", "how do I use this?", "how does this work?", "where do I start?", "怎么用", "说明书"…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Add a ControlNet branch to enforce spatial and structural constraints that text prompts alone cannot express. Use when the verifier reports flat background, wrong 3D layout, blurry edges, incorrect pose, poor surface texture, or when fixstrategy contains "addcontrolnet". ControlNet is the right tool whenever structure…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Strategy for prompts that contain exact object counts, multi-subject conjunctions, or unusual attribute bindings. Use when the prompt has a number word ("two", "seven", "exactly N"), joins subjects with "and" / "|" / "+", pairs a color or material with an unlikely noun ("green croissants", "metal toys", "wooden…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Set sampler parameters and prompt tags for vivid, concept-art-style outputs. Use when the user asks for "creative", "artistic", "fantasy", "concept art", "surreal", "stylized", or any clearly fictional scene where photorealism would be wrong. Pair with prompt-artist to also enrich the prompt text — this skill handles…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Configuration guide for DreamShaper 8 LCM (Latency Consistency Model). MUST be read before tuning any sampler parameter when the active checkpoint contains "dreamshaper" and "lcm" in its filename. LCM models use a completely different sampler, step count, and CFG range than standard diffusion models — applying…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Raise output fidelity by tuning prompt tokens, sampler parameters, and resolution. Use when the user asks for "high quality", "detailed", "sharp", "crisp", "professional", or "8K" — or when the verifier reports soft detail, noise, or underwhelming fidelity. Layer this on top of photorealistic or creative; the…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Add a second-pass upscale-and-refine stage to recover fine detail lost when the base latent is small. Use when the verifier reports blurry output, soft texture, or lack of fine detail — especially when the base latent is 768×768 or smaller. Also trigger when fixstrategy contains "addhiresfix". This is the most…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Edit a specific region of an existing image while leaving the rest untouched. Use when the user wants to "inpaint", "mask", "edit just the X", "remove the Y", "replace the background", "change only the face/hands/sky", or asks to fix a localized defect (extra fingers, weird eyes, dirty patch) without regenerating the…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Insert LoRA adapter nodes to correct defects the base model cannot fix through prompting alone — plasticky texture, flat lighting, distorted anatomy, style inconsistency, or soft overall quality. Use when the verifier reports any of these issues or when fixstrategy contains "injectlora". Always query available LoRAs…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Detect missing ComfyUI model weights and request user-approved downloads into the correct models subdirectory.
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Construct effective negative prompts for diffusion models. Use when the workflow has an empty or weak negative encoder, when the verifier reports defects the positive prompt can't fix (extra limbs, watermarks, low quality, wrong style bleed), when the user complains about "too cartoon", "too anime", "blurry"…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Tune prompts and sampler settings to produce camera-realistic images with natural lighting and DSLR-like quality. Use when the user describes a real-world scene or uses words like "photo", "photograph", "realistic", "DSLR", "cinematic", "RAW photo", or "shot on [camera]". Trigger this even when the user just says…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Rewrite a flat or sparse prompt into vivid, multi-dimensional artistic language. Use when the user asks for "creative", "aesthetic", "dreamy", "artistic", "masterpiece", "professional art", "award-winning", "concept art", "surreal", "futuristic", or "imaginative" images — or whenever the prompt feels too plain and the…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Configuration guide for Qwen-Image-2512, Alibaba's top-ranked open-source text-to-image model (Dec 2025). Detect this model when the workflow contains UNETLoader with "qwenimage" in the model name, or CLIPLoader with type "qwenimage". Uses standard ComfyUI nodes (UNETLoader, CLIPLoader, VAELoader, KSampler…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Apply separate text prompts to distinct image regions to prevent style contamination between subject and background. Use when the verifier reports subject and background look mismatched, scene elements compete stylistically, background is too plain or generic relative to the subject, or fixstrategy contains…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Choose the right KSampler samplername + scheduler pair for the active checkpoint. Consult before changing sampler / scheduler / step count, when the verifier reports overcooked detail, grainy noise, color cast, or posterized output, or when a workflow defaults to dpmpp2m + karras and the user wants faster / sharper /…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.
Moms-Organic-Agent-Lab/comfyclaw
Agent
Analyze blind comparison results to understand WHY the winner won and generate improvement suggestions.
Moms-Organic-Agent-Lab/comfyclaw
Agent
Compare two outputs WITHOUT knowing which skill produced them.
Moms-Organic-Agent-Lab/comfyclaw
Agent
Evaluate expectations against an execution transcript and outputs.
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Rewrite prompts to explicitly encode spatial relationships and physical arrangements between objects. Use when the user describes scenes with multiple elements using words like "on the left/right", "above/below", "behind/in front of", "jumping over", "in a circle", "partially hidden", or when multiple characters or…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Restructure prompts for accurate in-image text rendering. Use when the user wants specific text, quotes, signs, labels, or logos to appear in the generated image — including phrases like "a sign saying...", "text on the shirt", "a logo with the word", or whenever the user places target text in quotation marks. Text in…
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Architecture recipes for building ComfyUI video-generation workflows from scratch. Activate whenever modality=video and the workflow is empty or near-empty. Pairs with model-specific skills like wan-video-2.2 for parameter values; this skill owns the graph topology.
Moms-Organic-Agent-Lab/comfyclaw
Skill Claude CodeCodex
Configuration guide for Wan2.2, Alibaba's open-source text-to-video and image-to-video model. Activate when the user wants video generation (modality=video) and the available models include any of "wan", "wan2.2", or "wanvideo". Covers both ComfyUI native Wan nodes and the kijai WanVideoWrapper custom-node ecosystem …