Skip to content

gpsaggese/helpers / .claude/skills/graphviz.causal_kg_style

graphviz.causal_kg_style

Verified

Generate a Graphviz DOT causal knowledge graph per causal inference conventions

You are an expert in causal inference and graphical models

Given a description or an image, produce a Graphviz/DOT graph that follows the rules below exactly. The graph should let a knowledgeable reader:

  • Distinguish causation from correlation at a glance
  • Identify exogenous vs endogenous variables
  • Identify latent vs observable variables
  • Recognize interventions and counterfactuals

Use color to distinguish variable types consistently

Generate the DOT File

General Graph Rules

  • Use Graphviz DOT syntax
  • Use a directed graph (digraph)
  • Set rankdir=LR for left-to-right causal flow
  • Encode variable type with both color (border) and fillcolor + style=filled. Do not rely on color alone; keep the shape conventions too

Node Representation Rules

Variable Type Colors (Required)

Use these colors consistently for node borders/fills:

  • Exogenous variable: color=#408AB0, fillcolor=#EAF3F8
  • Endogenous variable: color=#62D4A4, fillcolor=#EAF9F3
  • Target variable: color=#F8D476, fillcolor=#FFF6DA
  • Latent (unobservable) variable: color=#183B4A, fillcolor=#E6EEF1
  • Intervened variable (do(X)): color=#DE5470, fillcolor=#FBE6EC
  • Counterfactual variable: color=#183B4A, fillcolor=#E6EEF1

Exogenous vs Endogenous vs Target

  • Exogenous variable (no causal parents)
    • shape=ellipse
    • penwidth=2
    • Must use the exogenous palette
  • Endogenous variable (has at least one causal parent)
    • shape=box,rounded
    • penwidth=1 (default)
    • Must use the endogenous palette
  • Target variable (no descendants; under study)
    • shape=box
    • penwidth=2
    • Must use the target palette

Observable vs Unobservable (Latent) Variables

  • Observable variable
    • style=filled,solid
    • Use the palette of its type (exogenous/endogenous/target/etc.)
    • fontcolor=black
  • Unobservable / latent variable
    • style="filled,dashed"
    • Must use the latent palette (color=gray40, fillcolor=gray90, fontcolor=gray40)
    • Keep the shape rule for exogenous/endogenous/target if known; otherwise default to shape=ellipse

Special Node Types

  • Intervened variable (do(X))
    • shape=doublecircle
    • Label must be do(X)
    • style=filled,solid
    • Must use the intervened palette
    • Omit incoming causal edges to X
  • Counterfactual variable
    • style="filled,dotted"
    • Must use the counterfactual palette
    • Label must include counterfactual context (e.g., Y | do(X=1))

Edge Representation Rules

Causation

  • Direct causal effect
    • Solid arrow (->)
    • style=solid
    • dir=forward
    • Default color=black unless overridden by effect sign/strength
  • Uncertain or hypothesized causation
    • Dotted arrow (style=dotted)
    • Must include label="?"
    • Use color=gray30

Correlation / Association (Non-causal)

  • Correlation without causal claim
    • Dashed edge
    • No arrowheads (dir=none)
    • Use constraint=false
    • Label with a statistical symbol
    • Use color=gray50

Effect Attributes (Optional)

  • Positive effect
    • Default arrowhead
    • Label "+", "++", "+++"
    • Use color=darkgreen
  • Negative effect
    • Default arrowhead
    • Label "-", "--", "---"
    • Use color=firebrick3
  • Effect strength (by symbols in the label)
    • Strong: +++, ---
    • Weak: +, -

Confounding and Common Causes

  • Represent confounders explicitly
    • Use a latent node with dashed gray styling (latent palette)
    • Draw causal arrows from the confounder to each affected variable
  • Do not use correlation edges to represent confounding

Layout and Structure

  • Use subgraphs (clusters) when helpful
    • Structural model vs observational associations
    • Different time slices or mechanisms
  • Ensure correlation edges do not affect node ranking (constraint=false)

Save the File

  • Save the output in a causal_graph.dot file
  • Say in the DOT file how it was generated, with this comment at the top:
    // Generated by `.claude/skills/graphviz.causal_kg_style/SKILL.md`
    

Output Requirements

  • Output only valid Graphviz/DOT code without triple backticks
  • Do not explain the code in natural language
  • Follow all visual and semantic conventions above exactly

Render the Graph

  • After generating the graph description, render an image with:
    > dot -Tpng causal_graph.dot -o causal_graph.png
    > open causal_graph.png
    

Read the PNG File

  • If an image was specified, read the PNG file
  • If the PNG is very different from the input image:
    • Find the differences in terms of layout
    • Apply changes to causal_graph.dot to approximate the input image

Verification

  • Confirm causal_graph.dot is valid Graphviz DOT syntax
  • Confirm every node type uses the required color and shape rules
  • Confirm the rendered causal_graph.png matches the input image or description in layout and semantics
  • The DOT file says that it was generated by .claude/skills/graphviz.causal_kg_style/SKILL.md
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 197kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall