Skill rating

Model a building, object or set in a 3D app (Blender, Maya or Houdini) so that it matches a reference photo — the way a 3D artist does it - measure the photo, match the camera, block out grey shapes u

Resize images to a specific resolution OR downsize to fit Claude API upload limits (max 1568px long edge, max 5MB). Use the exact-resize mode when the user requests a specific width/height; use downsi

Build a hook pipeline on the user's canvas yourself — stages, reviews, loops, parallel branches — out of agentY's own nodes, wired with the execution wire. Activate when the user asks you to set up, d

Run the workflow the user already has open in the ComfyUI canvas over and over, judging each output against a condition they stated and changing one value until it is met — the closed "keep trying unt

Destroy a 3D model in Houdini with a rigid-body (RBD) simulation — a building, a set or an object collapsing, breaking apart, being knocked over — plus dust and simple cloth, rendered as an OpenGL fli

Build a ComfyUI workflow from scratch when no template matches (template.name == "build_new"). Builds to the standard in the recipe database (task -> model -> node clusters), starting from a recipe me

Make EDITABLE motion graphics - titles, lower thirds, logo reveals, animated text / shapes / pictures - by writing a scene file (*.mgscene.json, data, not React), checking it, looking at stills, rende

Ultimate SD upscaler on Flux1-dev fp8 (template upscale_ultimateSD). Activate in the Query Templates when the selected template is upscale_ultimateSD. Activate in the Assemble Workflow during assembly

Turn a whole storyline into a production blueprint for a short film: a story bible plus the entire story split into Kling multi-shot SEQUENCES of <=10s each, ending with a single machine-readable JSON

Kling 3.0 multi-shot storyboard (template Kling3_multiShot). Activate in the Query Templates when the selected template is Kling3_multiShot — overrides the standard prompting step. Activate in the Ass

When and how to delegate a job to a fresh, specialized subagent — plan first, then spawn an executor with a MINIMAL toolset and a single skill. Use for complex/multi-step/batch jobs, or anything that

Write a very short synopsis (a logline of a few words up to ~2 short sentences) capturing a storyline's premise. Activate whenever the user wants a new story idea, logline, premise, or storyline seed.

Make the user's open canvas readable - arrange nodes so none overlap, split the graph into chunks, draw and name groups, colour them consistently, and save the workflow. Activate when the user says th

ModelSamplingFlux node patch requirements. Activate in Brain step 2 (Patch and validate) whenever the workflow contains a ModelSamplingFlux node.

Train a LoRA for an image or video model - a person, pet, product, style or motion - with the agent doing every preparation step and the USER approving the dataset before anything trains. Use for "tra

Analyse video input using Gemini, create motion prompts. Activate when brainbriefing template name contains 'video_gemini_motionPromptGeneration'. Patchable: Input video, Output; Locked: Structural no

Expand a synopsis into a sequence of consistent, visual scene descriptions that serve as the starting point for start-frame image and video generation. Activate when the user provides (or you just pro

Install and set up an MCP server completely and deterministically - a catalog preset in one call, or a server that is not in the catalog through a recipe you draft from its README and the user approve

Make programmatic motion-graphics video (titles, promos, explainers, animated text/data, stills, GIFs) with Remotion — React compositions rendered to mp4. Activate when the user asks for a Remotion vi

Turn a model's GitHub repository (already cloned locally) into a self-contained, importable ComfyUI custom-node pack — __init__.py, nodes.py, requirements.txt, README.md, pyproject.toml. Use when hand

Make a new shot with a chosen camera move from ONE picture - picture to 3D scene, animate a camera through it in the bEpic viewer, render the move, and use the render as the control video of a video m

Generate an image or video with an Alibaba DashScope (Model Studio) model - Wan, HappyHorse, Qwen-Image - billed to the user's own DashScope key instead of ComfyUI partner credits. Activate when the u

Make a 3D scene from a picture or a text prompt with World Labs Atlas (a depth mesh in the picture's camera), grow it with new views, use it as the set for render_camera_move, and render single new vi

Load files into the bEpic image viewer when the user asks to see them there - "show me clip 3 in the viewer", "open these frames side by side", "open the edit", "put the two renders in a tab". Also op

Step-by-step procedure for handing off multi-iteration batch workflows to the Executor. Activate when count_iter > 1, or when applying the SAME workflow to multiple input images (Mode C: swap the inpu

Render a ComfyUI workflow on the studio's Deadline render farm instead of the local ComfyUI. Use for "render this on the farm", "send it to Deadline", "run 20 variations on the farm", "is the farm fre

Use ONLY when count_iter > 1 AND variations is true in the brainbriefing. Generate N distinct variation prompts and write them to multiprompt.json so the pipeline can split them across individual work

Handles user-provided annotated images. When the user submits a drawing or mark-up on top of an image alongside a message, this skill wires the last generated output as the primary edit target and the

Run ONE partner (API) node directly - Seedance, Kling, GPT Image, Nano Banana / Gemini image, Ideogram, Meshy, Veo, Runway, Hailuo ... - with partner_run, no template search. Activate when the request

How to extend yourself — capture a reusable procedure as a skill (create_skill), spawn a subagent for a heavy self-contained job (spawn_subagent), or author a ComfyUI node pack from a model's GitHub r

Media-kind output_path mapping for brainbriefing output_nodes. Activate during step 6 (Identify output nodes) to set the correct output_path for each output node.

Consult long-term memory before assembling or patching a ComfyUI workflow, and especially before retrying a failing pattern (repeated tool calls to fix the same assembly sub-problem, or the same valid

Select ComfyUI workflow templates based on user requests. Retrieves metadata about the workflow, such as used models and input / output nodes, to assist with workflow assembly.

Assembles a workflow on the basis of a brainbriefing JSON and a pre-selected template.

Recipe-based ComfyUI workflow building — the (task -> model -> node clusters) knowledge base. Front door for building a workflow from a recipe; the full build procedure lives in the assemble-new-workf

Where the open-weight models of a run execute - this machine's GPU or Comfy Cloud. Use for "run this in the cloud", "use Comfy Cloud", "keep it local", "run it on my GPU", "why did that run in the clo

ComfyUI prompt engineering — universal CLIP/weight syntax plus per-model prompt rules loaded on demand from references/. Activate whenever composing a generation or edit prompt.

Seedance 2.0/2.5 node wiring (ByteDance2ReferenceNode, ByteDance2TextToVideoNode, ByteDance2FirstLastFrameNode). Activate in the Query Templates when a Seedance template is selected, and in the Assemb