Notes

The presenter's walk sequence through the slide. Not a script — a map.

Must contain:

• Walk sequence: which visual element to reference first, second, third
• Key phrases: exact words the presenter should say (quoted)
• Named concepts: systems, frameworks, or terminology to name explicitly
• Audience connection: one sentence linking the slide to audience concerns

Prohibited:

• Generic filler ("This slide shows..." or "As you can see...")
• Repeating the slide title as the opening line
• Instructions to the presenter ("Click to advance")

Field 2: talking_points (4-8 speakable bullets)

Bullets the presenter can read naturally. Each is one breath — 10-20 words.

Must contain:

• 4-8 speakable bullets covering the slide's content
• 1-2 objection handlers or anticipated questions (prefixed with "If asked:")
• At least one transition phrase connecting to the next slide

Prohibited:

• Bullets that duplicate slide text verbatim
• More than 8 bullets (forces the presenter to rush)
• Abstract bullets without concrete details

Field 3: Template-Configurable

The third field's name and content type are read from the active workflow template's notes.tab3 configuration. If no template is active, defaults to references.

Notes JSON Format

Notes are stored alongside slides in notes.json:

{
  "slides": [
    {
      "slide_index": 0,
      "slide_file": "01-ai-powered-analytics.html",
      "notes": "Start with the headline number — say 'sixty percent faster decisions.' Point to the three stat cards left to right: cycle time, insight speed, adoption. Pause on adoption — this is the surprise number that proves it's not just a tool upgrade.",
      "talking_points": [
        "Our AI platform cut decision latency from 14 days to under 6",
        "That's not a lab number — it's measured across 200 active users",
        "Adoption hit 92% in the first quarter without a mandate",
        "If asked: 'What about the remaining 8%?' — They're in departments still onboarding, not resistant users",
        "This sets up the expansion case we'll cover next"
      ],
      "references": [
        "Decision cycle data: Q4 2025 Operations Dashboard, p.12",
        "Adoption survey: Internal UX Research, January 2026"
      ]
    }
  ],
  "template": "general",
  "tab3_field": "references",
  "generated_at": "2026-03-23T10:00:00Z"
}

Batch Mode

When processing 4 or more slides, batch mode activates automatically.

Dispatch Strategy

1. Read all slide HTML files and any existing notes.json
2. Group slides into batches of 3
3. Dispatch each batch as a parallel subagent
4. Each subagent receives: the slide HTML content, the outline JSON (for narrative context), and any existing notes to preserve
5. Collect results, validate against the quality gate, merge into notes.json

Why Groups of 3

• Small enough for focused attention per slide
• Large enough to maintain narrative continuity between adjacent slides
• Overlapping context: batch N gets the last slide of batch N-1 for transition awareness

Preserving Existing Notes

Before overwriting, read existing notes.json. For each slide:

• If notes exist and are marked "user_edited": true, preserve them entirely
• If notes exist but are auto-generated, upgrade them
• If no notes exist, generate fresh

Quality Gate Checklist

Every generated note must pass these checks before writing:

• notes field is 200-500 characters (not shorter, not longer)
• notes field references specific visual elements on the slide
• notes field does not start with the slide title
• talking_points has 4-8 bullets
• At least one bullet starts with "If asked:" (objection handler)
• At least one bullet contains a transition to the next slide
• No bullet duplicates slide text verbatim
• Tab 3 field is populated with content matching its type
• No generic filler phrases present ("This slide shows", "As you can see", "Let me walk you through", "Moving on to")

If any check fails, revise that slide's notes before writing.

Integration with Other Skills

Reads From

Produces For

Usage

# Generate notes for all slides in a directory
python ${CLAUDE_SKILL_DIR}/scripts/slide_notes.py \
  --slides-dir ./slides \
  --outline outline.json \
  --output notes.json

# Upgrade existing notes (preserves user edits)
python ${CLAUDE_SKILL_DIR}/scripts/slide_notes.py \
  --slides-dir ./slides \
  --existing notes.json \
  --upgrade \
  --output notes.json

# Generate notes for a single slide
python ${CLAUDE_SKILL_DIR}/scripts/slide_notes.py \
  --slide ./slides/03-key-metrics.html \
  --outline outline.json \
  --output notes.json

# Override tab3 field name
python ${CLAUDE_SKILL_DIR}/scripts/slide_notes.py \
  --slides-dir ./slides \
  --tab3-field learning_objectives \
  --output notes.json

How Claude Should Use This Skill

1. Read the outline first — notes quality depends on understanding each slide's purpose in the narrative arc, not just its visible content.
2. Read existing notes — never overwrite without checking for user edits.
3. Write notes in slide order — transitions depend on knowing the next slide.
4. Run the quality gate — every note, every time. No exceptions.
5. Ask about tab3 — if no template is active, ask the user what the third field should contain before generating.