Keynote Slides Skill | Agent Skills

Keynote Slides

Assets

assets/keynote-slides.html holds the single-file slide deck template.
references/brand-guidelines.md captures brand tokens, typography, and image style guidance.
references/gemini-media.md documents the Gemini nano banana and Veo media settings.

Workflow

Resolve entity + detect content-db. Read deck.json → entity field; fall back to <section data-entity="..."> on the active slide; then to brands.js default entity (the entity whose export key is "default", or the single entity in the file). Check whether content-db/<entity>/ exists alongside brands.js. If it does and is well-formed, content-db mode is active — load relevant atoms before generating any content. If it does not exist or is malformed, proceed normally with no message. See the Content Database section for the full protocol.
Run the deck bootstrap to create a deck folder:

scripts/new-deck.sh example-pitch --entity northwind --title "Example Pitch" --type pitch

Update decks/brands.js when brand tokens change.
Edit decks/<deck-id>/index.html and duplicate slides inside <main id="deck">, keeping each data-title unique.
Use layout classes (layout-title, layout-split, layout-grid, layout-metrics, layout-quote) to keep spacing consistent.
Apply reveal plus --reveal-index to stagger key elements.

Entities

Use the generator panel to select the active entity profile.
Add data-entity="entity-id" on a slide to override the global profile for that slide.
Add ?entity=entity-id to the URL for a quick switch.
Use mediaPromptPrefix in brandProfiles to keep Gemini media outputs on brand.

Deck storage

decks/<deck-id>/index.html is the editable deck file.
decks/<deck-id>/deck-config.js stores deck metadata (entity, title, resources).
decks/<deck-id>/deck.json stores the same metadata in JSON form.
decks/<deck-id>/slides.md is for draft copy and notes.
decks/<deck-id>/resources/assets/ holds logos, images, and media inputs.
decks/<deck-id>/resources/materials/ holds briefs, pricing docs, P&L inputs, and outlines.

PMF Panel Integration

Export PMF Panel findings as deck input materials:

# From pmf-panel directory
.claude/export-deck-brief.sh <scenario-id>

# Copy to deck materials
cp scenarios/<scenario-id>/deck-brief.md \
   ../keynote-slides-skill/decks/<deck-id>/resources/materials/pmf-brief.md

The brief includes diagnosis, scenarios, 30/60/90 plan, and specialist perspectives ready for narrative engine ingestion.

Collaboration

Co-author the narrative: propose headlines, POV, and slide ordering based on deckType and entity preferences.
Keep the brief in resources/materials/brief.md and capture evolving preferences in deck.json or decks/brands.js.
Use concise headline options (3-5 variants) and confirm direction before building slides.

Review loop

Use the Chrome Devtools MCP tools to capture a snapshot/screenshot and review layout.
Check hierarchy, alignment, spacing rhythm, and contrast; then adjust copy and spacing.
Use the generator panel for brand-aware media, then re-check balance and whitespace.
Register new content-db atoms (if active). After review, write new atoms to content-db/<entity>/ for any claims (cl###), copy strings (cp###), or assets (as###) introduced during the build. Run node content-db/validate.js (or equivalent). The output package is not complete until validation exits 0. Do not modify existing atoms — add a vl### validation record and mark the original status: disputed if wording conflicts.

Templates

Copy markup from the <template> blocks at the bottom of the file.
Replace placeholders with branded copy, numbers, and visuals.

Media generation

Add data-gen and data-prompt to <img> or <video> elements.
Open the generator panel with g or the Gen button.
Save the API key and model settings to localStorage (never commit keys).
For image-to-image or image-to-video, load a base image in the panel.
Run "Generate slide" or "Generate all" and review outputs.

Model-Mediated Image Acquisition

Use the /acquire-images skill (see skills/acquire-images.md) to populate slides with visuals.

Claude decides whether to generate (Gemini) or search (stock photos) for each slide:

| Content Type | Decision | Why | |--------------|----------|-----| | Diagrams, flowcharts | GENERATE | Custom layouts, brand colors | | Data visualizations | GENERATE | Precise data representation | | Real-world photos | SEARCH | Authentic people, places | | Team/people shots | SEARCH | Realistic human photos | | Abstract concepts | GENERATE | Metaphorical, brand-styled | | Branded hero images | HYBRID | Search base + AI overlay |

Search sources: Unsplash, Pexels, Google Custom Search.

Attribution: Downloaded images tracked in resources/materials/image-credits.json.

See skills/acquire-images.md for the full workflow.

Preview

scripts/serve-decks.sh

Then open http://<tailscale-ip>:8921/decks/<deck-id>/index.html.

Speaker notes

Add per-slide notes with a hidden block:

<aside class="slide-notes">Speaker notes here.</aside>

Toggle the notes panel with the "Notes" button or press n.
Append ?notes=1 to open notes by default.
Use "Export notes" to download a markdown file.

Animation + SVG

Use data-anim for lightweight entrance animations (fade, slide-up, slide-left, slide-right, scale-in).
Set --anim-delay to stagger; avoid mixing with reveal on the same element.
Disable animation with ?motion=off or rely on prefers-reduced-motion.

Inline SVG diagrams use .diagram and data-media="svg":

<svg class="diagram" data-media="svg" viewBox="0 0 800 450" role="img" aria-label="Diagram"></svg>

Keep media lanes explicit:
- data-gen = Gemini only (optional data-media="gemini").
- Inline SVG = no data-gen.
- Static images/videos = no data-gen.

Micro Animations

Micro animations are small decorative accent elements that add subtle motion to slides without distracting from content.

What micro animations ARE:

Small accent lines that expand/draw next to headlines
Subtle decorative strokes or shapes that pulse gently
Tiny visual flourishes positioned near text (not on text)
CSS-driven, lightweight, and unobtrusive

What micro animations are NOT:

Large SVG graphics or overlays covering slide areas
Animations applied directly to text (no flashing, glowing, or bouncing text)
Complex particle systems or heavy motion graphics
Anything that competes with or obscures content

Implementation:

<!-- Accent line after a headline -->
<h1 class="title">Your headline here</h1>
<span class="accent-line"></span>

Available classes:

.accent-line — 80px amber line that expands from left, then pulses
.accent-line.long — 120px version for major headlines

Behavior:

Line draws in from left (0.8s ease-out) when slide becomes active
Then gently pulses (opacity + slight scale) on a 4s cycle
Respects prefers-reduced-motion and ?motion=off

When to use:

Title slides (layout-title) to add visual interest
Key message slides where you want emphasis
Sparingly — 3-4 slides per deck maximum

Copy editor

Open decks/<deck-id>/editor.html in a second window to edit copy without touching HTML.
Use the editor "Open deck" button to connect and update the live preview.

For existing decks, copy the template first:

cp skills/keynote-slides/assets/keynote-editor.html decks/<deck-id>/editor.html

Edits are stored in localStorage; export JSON from the editor for handoff.

PDF export

Use the browser print dialog and "Save as PDF".
Enable background graphics for gradients and color fills.
The template includes print styles to paginate each slide.

CLI option:

node scripts/export-pdf.js decks/<deck-id> --out /tmp/<deck-id>.pdf

Navigation

Arrow keys, PageUp/PageDown, Space.
Home/End for first or last slide.
Use #slide-title hash navigation for direct jumps.

Review Mode / Feedback System

Enable reviewers to leave comments on deck elements for collaborative feedback.

Entering Review Mode

Click the "Review" button in the bottom toolbar (next to Gen)
Press r key to toggle review mode
Add ?review=1 to the URL to start in review mode

Adding Comments

In review mode, hover over elements to see them highlighted
Click any commentable element (titles, text, cards, metrics, media frames)
Or select text within an element to comment on specific wording/typos
First-time commenters enter name and email (stored in session)
Type feedback in the popover and click "Add Comment"
A numbered badge (①②③) appears on commented elements for easy reference

Viewing Comments

Press c or click the sidebar toggle to open the comment sidebar
Comments are grouped by slide
Click "Go to slide" to navigate to the commented element
Mark comments as resolved or delete them

Exporting Feedback

From the comment sidebar:

Export JSON - Downloads comments-<deck-id>.json for backup or import
Export MD - Downloads markdown summary for sharing or Claude iteration

Feedback Viewer Page

Use assets/feedback-viewer.html to review all feedback outside the deck:

Open feedback-viewer.html in a browser
Load a comments.json file or enter a URL
Filter by open/resolved status
Mark comments as resolved
Export updated JSON or markdown

Add ?url=<path-to-json> to auto-load comments.

Comment Data Structure

Comments are stored in localStorage keyed by deck ID. Export structure:

{
  "deckId": "example-pitch",
  "nextNumber": 4,
  "comments": [
    {
      "id": "c_1706123456789_abc123",
      "number": 1,
      "slideIndex": 2,
      "slideTitle": "Our Solution",
      "elementSelector": "[data-comment-target='headline']",
      "elementText": "First 50 chars of element...",
      "selectedText": "specific phrase",
      "comment": "This needs more specificity",
      "author": {
        "name": "Sarah Chen",
        "email": "sarah@example.com"
      },
      "createdAt": "2024-01-24T10:30:00Z",
      "resolved": false
    }
  ]
}

number: Sequential comment number (①②③) for easy reference in feedback
selectedText: If reviewer selected specific text, captures that selection (null otherwise)

Element Targeting

For precise comment targeting, add data-comment-target attributes:

<h2 data-comment-target="solution-headline">Our Solution</h2>
<p data-comment-target="value-prop-1">We reduce costs by 40%...</p>

Elements without explicit targets use a generated CSS selector path.

Narrative Engine Integration

For content-driven deck creation, use the Narrative Engine workflow that matches your material to proven storytelling frameworks.

Reference Files

references/narrative-engine/narrative-arcs.md — Beat-by-beat structures for 10 narrative arcs
references/narrative-engine/framework-selection.md — Selection matrix by audience/purpose/content
references/narrative-engine/framework_selection_guide.md — Deep pairing guidance for arcs + frameworks
references/narrative-engine/communication-frameworks.md — 7 efficiency-optimized frameworks
references/narrative-engine/checklists.md — Quality gates for narrative + copy review
references/narrative-engine/agent-reference-*.md — Agent-specific frameworks for review

Workflow: Narrative Build

Ingest resources: Run node scripts/ingest-resources.js decks/<deck-id> to read all materials
- Or use node scripts/narrative-build.js decks/<deck-id> to prepare model-mediated prompts 1a. Load content-db atoms (if active). If content-db mode is active, pull copy.md and claims.md atoms for the resolved entity. Filter atoms by audience tag if present. Surface any atom with a non-null gate value before proceeding — do not include gated content without user confirmation. If content-db is not active, skip this step.
Focal discovery + discovery: Align on the one point, then answer 5 questions (audience, purpose, content type, tone, reveal)
Density + framework match: Choose density mode, then get 2-3 recommendations with content mapped to structure
Deck generation: Build slides with source attribution tags. If content-db is active: use each claim's exact approved text — do not paraphrase. For any claim or copy string not found in the db, pause and ask: "That claim isn't in content-db/<entity>/claims.md — want me to register it as unverified, or use the closest approved atom?" Do not silently include or skip unapproved content.
Review panel: 5 agents + Director synthesize feedback

Discovery Questions

| Question | Options | |----------|---------| | Audience | Executive, Technical, Investors, Skeptics, General, Mixed | | Purpose | Persuade, Inform, Inspire, Align, Report, Defend, Entertain | | Content type | Research, Strategy, Origin story, Post-mortem, Pattern insight, etc. | | Tone | Authoritative, Provocative, Warm, Urgent, Balanced, Visionary | | Reveal potential | Yes (has surprise), No (straightforward), Help me find one |

Framework Selection Quick Reference

| If your content has... | Consider... | |------------------------|-------------| | A genuine surprise | The Prestige or Mystery Box | | Multiple stakeholder views | Rashomon | | A transformation story | Hero's Journey | | Future vision | Time Machine | | Root cause analysis | Columbo | | Strategy/roadmap | The Heist | | Paradigm shift | Trojan Horse |

5-Agent Review Panel

| Agent | Lens | Key Question | |-------|------|--------------| | Audience Advocate | Target audience persona | "Does this land for [audience]?" | | Comms Specialist | Messaging, emotion, PR risk | "Is this tight and bulletproof?" | | Visual Designer | Metaphor coherence, S.T.A.R. moments | "What visual makes this unforgettable?" | | Critic | Pacing, weak links, efficacy | "What's the weakest link?" | | Content Expert | Accuracy, logic, sources | "Can every claim be defended?" |

Stress Test Panel (Optional)

After the 5-agent review, optionally stress-test with stakeholder personas auto-selected by content type:

| Persona | Questions | Best For | |---------|-----------|----------| | Engineer | "How does this actually work?" | Technical proposals, product launches | | Skeptic | "Why should I believe this?" | Bold claims, paradigm shifts | | Risk Officer | "What could go wrong?" | Strategy, transformation, investment | | CFO | "What are the numbers?" | Pitches, business cases, ROI claims | | Lawyer | "What's the exposure?" | Policy, compliance, external-facing | | Conservative | "Why change what's working?" | Change management | | COO | "Would this actually work?" | Execution plans, go-to-market |

The Director triages findings into Must Fix, Should Fix, and Could Fix categories.

Source Attribution Tags

| Tag | Meaning | |-----|---------| | [DIRECT] | Verbatim from source material | | [PARAPHRASE] | Restated ideas | | [ELABORATED] | Expanded concept | | [SYNTHESIZED] | Combined multiple sources | | [GENERATED] | New content for flow |

Headline Rules

Image & Action: Concrete nouns + strong transitive verbs; avoid "is/are"
Tension & Turn: Because/Therefore, Not/But, Before/After
Cadence: 8-14 words; two-beat rhythm
Specific Anchors: Time/place/actor/number in every third headline
Power verbs: tilts, unseats, ignites, drains, compounds, unlocks, anchors, accelerates
Metaphor family: One per deck (journey OR ecology OR weather, etc.)

See /docs/integrated-architecture.md for full technical details.

Content Database

A modular, approved-content management layer for any project using keynote-slides. When a content-db/<entity>/ directory exists alongside brands.js, agents automatically load and enforce approved atoms. When it does not exist, the skill runs exactly as before — no config, no warnings.

Directory Structure

<deck-repo>/
├── brands.js                  ← entity source of truth (unchanged)
├── content-db/
│   └── <entity>/              ← one directory per entity
│       ├── README.md          ← schema reference + agent protocol
│       ├── claims.md          ← cl### atoms — stats, benchmarks, data points
│       ├── validation.md      ← vl### atoms — experimental evidence blocks
│       ├── assets.md          ← as### atoms — images, video, SVG with provenance
│       ├── copy.md            ← cp### atoms — approved text passages
│       ├── brand.md           ← br### atoms — color tokens, typography
│       └── layouts.md         ← ly### atoms — approved slide structures
└── <deck-id>/                 ← decks alongside (unchanged)

content-db/ always lives next to brands.js. Multi-entity repos have one subdirectory per entity.

Entity Resolution

Resolve the active entity in this order:

deck.json → entity field
<section data-entity="..."> on the active slide
brands.js → the entity whose export key is "default", or the single entity in the file. If multiple entities exist without a "default" key, surface the ambiguity — do not guess.

Detection

resolve entity
→ content-db/<entity>/ exists and is well-formed?
  → YES: load relevant atom files, activate compliance
  → MALFORMED: log warning, proceed without compliance
  → NO: proceed as normal, no message

| Operation | Atom files loaded | |-----------|------------------| | Narrative planning | copy.md, claims.md | | Full build | all six files | | Targeted edit | only types relevant to the change |

Compliance Rules

Read before write. Load relevant atoms before generating content. Never re-derive brand colors or claim text if an atom already exists.
Surface unapproved content (Build and Edit only — not planning). If a claim or copy string is not in the db (different number, source, or framing = new claim; minor rewording = same), pause: "That claim isn't in content-db/<entity>/claims.md — want me to register it as unverified, or use the closest approved atom?"
Register new atoms. After any build or edit, write atoms for all claims (cl###), copy strings (cp###), and assets (as###) introduced. Run node content-db/validate.js. Output is not complete until exit 0.
Never modify existing atoms. If content conflicts with an existing atom, add a vl### validation record and update status: disputed in the original. Surface the conflict to the user.

Atom Prefixes

All new atoms use a two-letter prefix + zero-padded three-digit number. Existing atoms with single-letter prefixes are grandfathered.

| File | Prefix | Contents | |------|--------|----------| | claims.md | cl### | Stats, benchmarks — status: validated \| disputed \| unverified | | validation.md | vl### | Experimental evidence — experiment_type, result, institution | | assets.md | as### | Images, video, SVG — file, deck, type, used_in | | copy.md | cp### | Approved text passages — concept, audience_level, tone, variants | | brand.md | br### | Color tokens, typography — element, css_var, value, usage_rules | | layouts.md | ly### | Slide structures — slide_type, css_classes, js_required, data_density |

Each atom is a ## <id> heading with bullet-field body. Full schema in content-db/<entity>/README.md.

Bootstrap

The skill does not scaffold a content-db unprompted. If a user asks why content-db is not active, explain the directory is absent and offer to create it. Only scaffold if the user says yes.

Bootstrap steps (when approved):

Create content-db/<entity>/ directory
Create six empty atom files with schema header comments
Create README.md with schema reference and compliance rules
Offer (separately) to extract initial atoms from existing deck HTML

If the user declines, proceed in no-compliance mode and do not raise the topic again in the same session.

Agent Skills: Keynote Slides

Install this agent skill to your local

Skill Files