analyze-comfyui-workflow-json
Analyze a ComfyUI workflow JSON and produce a structured map of nodes, links, subgraphs, and data flow before making changes. Use when user says "analyze this workflow", "map this comfyui json", "unde
SKILL.mdskills/analyze-comfyui-workflow-json/SKILL.md
GitHubProcess: Analyzing a ComfyUI Workflow JSON
1. Identify node types
Scan all "type" fields. Categorize into: input nodes (LoadImage, PrimitiveInt, etc.), processing nodes (custom UUIDs or named types), output nodes (SaveVideo, PreviewImage).
2. Map UUID subgraphs
UUID-typed nodes (e.g. "98e3bef8-...") are subgraphs. Find their definition in definitions.subgraphs[]. Read "name" field for human label.
3. Trace the link graph
"links" array format: [link_id, source_node_id, source_slot, target_node_id, target_slot, type]. Build a directed graph: source → target.
4. Identify data flow backbone
Follow primary data types (IMAGE, VIDEO, MODEL) through links. Ignore reroute nodes (type "Reroute") — treat them as pass-throughs, collapsing the chain.
5. Extract parameter nodes
PrimitiveInt, DF_Float, and similar nodes are constants. Read widgets_values[0] for their value. Read "title" for semantic meaning.
6. Find chaining pattern
Check if output slots (especially IMAGE) of one processing node feed as start_image input of another same-type node. If yes → sequential chaining pattern.
7. Reconstruct computation
For math nodes (multipliers, converters), read inputs labels and outputs labels. Infer operation from context (e.g. seconds × fps → frames).
8. Summarize architecture Express as: inputs → processing chain → outputs. Include parameter values, segment counts, and any detected patterns (chaining, branching, parallel paths).