Goal

• Given a markdown file or slides from the user <FILE>, propose visuals (e.g., diagrams, pictures, tables) that make the slides more intuitive

Workflow

Read Related Rules

• Read the file or the provided text
• Read .claude/skills/slides.rules.md for the slide conventions and rules
• Read .claude/skills/figure.rules.md for the visual rules

Propose a Visual for Each Slide

• If a slide has no visual element, consider what can illustrate its concepts
• Options, from ## Types of Illustrations in .claude/skills/figure.rules.md:
  • Table
  • Mermaid graph
  • Graphviz diagram
  • TikZ diagram
  • Images
  • Website screenshots

Output Example

• The proposal can look like:

  Slide 1: "A Map of Machine Learning"
  - **Current**: Has mermaid mindmap (present)
  - **Assessment**: Excellent visual already present. Keep as-is.
  
  Slide 2: "Machine Learning Paradigms"
  - **Proposed**: Comparison table showing Paradigm name, Question asked, and Key characteristic
    - Format: Use styled-table for consistency
    - Shows supervised vs unsupervised vs RL vs active learning side-by-side
    - Helps learners see the contrasts quickly
  
  Slide 3: "Machine Learning Theory"
  - **Current**: Text-only explanation of 4 theoretical foundations
  - **Proposed**: Graphviz diagram showing theoretical frameworks
    - Show how VC theory, Bias-variance decomposition, MDL, and Bayesian approach relate
    - Color by category: theoretical foundations
    - Add connections showing dependencies/relationships
  
  Slide 4: "Machine Learning Models"
  - **Current**: Text-only list of model variants
  - **Proposed**: Taxonomy diagram (Graphviz or TikZ)
    - Organize models by key distinctions: parametric vs non-parametric, linear vs non-linear
    - Show example models in each category
    - Helps learners understand the model space structure
  
  Slide 5: "Machine Learning Techniques"
  - **Current**: Text-only list of pipeline stages
  - **Proposed**: ML Pipeline flowchart (Graphviz)
    - Show the stages: Input Processing -> Model Building -> Performance Evaluation -> Diagnostic -> Regularization -> Aggregation
    - Color stages by function (data preparation, model, evaluation, improvement)
    - Include key techniques under each stage
    - This is the most impactful addition for learning
  
  Slide 6: "Some machine Learning Adages"
  - **Current**: Collection of quotes with attributions
  - **Assessment**: Text-based content is appropriate here. No visual needed.
  
  | Slide | Change                         | Type          | Impact    |
  |-------|--------------------------------|---------------|-----------|
  | 1     | Keep existing                  | -             | Good      |
  | 2     | Add paradigm comparison table  | Table         | High      |
  | 3     | Add theory framework diagram   | Graphviz      | Medium    |
  | 4     | Add model taxonomy diagram     | Graphviz/TikZ | High      |
  | 5     | Add ML pipeline flowchart      | Graphviz      | Very High |
  | 6     | Keep as-is                     | -             | Good      |
  

Save the Plan

• Save the plan to plan-slides.add_visuals.md
• Say in the plan file how it was generated, with this line right after the title:

  - Generated by `.claude/skills/slides.add_visuals/SKILL.md`
  

Ask User to Confirm

• Make a numbered list of the proposed changes for the user
• Once the user confirms, perform the changes

Constraints

• Maintain the structure of the text and keep the content of the existing text

Verification

• Make sure the updated slides work by running the flow, e.g., gen_slides.py

  > gen_slides.py -i book.Agentic_AI/12.1 --notes_to_pdf_args="--skip_action open_pdf"
  

• The plan file says it was generated by .claude/skills/slides.add_visuals/SKILL.md