OCI PPTX Skill

Oracle-Branded Presentations with Cognitive Science

Quick Reference

python -m markitdown file.pptx

Do NOT load multiple references simultaneously.

The Five Cognitive Science Rules

Based on Sweller (cognitive load theory), Mayer (multimedia learning), Phillips (Death by PowerPoint):

1. One Message Per Slide

Each slide = exactly ONE key point. Multiple messages = divided attention = 0% retention of secondary message.

2. Avoid Redundancy Effect

Reading text while hearing speech = 0% retention (Mayer). Max 50 words per slide. Move details to speaker notes.

3. Size = Importance Hierarchy

Largest element = most important content (may NOT be the headline). Make key metrics larger than titles when the metric IS the message.

4. Context-Aware Backgrounds

#0B2540

#1a1a1a

#FFFFFF

#F5F5F5

Dark slides make the SPEAKER the highest-contrast object in the room — that's the cognitive science intent.

5. Maximum 6 Objects Per Slide

Count ALL elements: images + shapes + text boxes + charts + decorative (exclude footer/header). Beyond 6: counting takes 500% more cognitive energy → audience disengages. Exception: Grid layouts where items form a single visual unit.

Oracle Brand Quick Reference

Typography

Font: Oracle Sans Tab (Regular/Bold)

• Verify installed:

  fc-list | grep -i "oracle"

• Fallback: Arial (always prefer Oracle Sans Tab)

Quick Colors by Pillar

For full palette details, read references/color-palettes.md.

#1E9773

#EECC44

#3E93B6

#1EB398

#26A3B1

#3FA67D

#D1798C

#5F9CC7

#9874B8

#5F9CC7

#8A8A8A

#C74634

Backgrounds: Dark

#0B2540

#1a1a1a

#FFFFFF

#F5F5F5

Oracle Footer (Mandatory on All Slides)

• Left:

  Copyright © 2026, Oracle and/or its affiliates

• Right (optional): omit for public events |

  | Confidential: Internal

  |

  | Confidential: Restricted

  |

  | Confidential: Highly Restricted

Design Workflow

Step 1: Context Assessment

1. Where presented? Large live event → Dark primary. Virtual → Mix. Document → Light.
2. Who is audience? Executives: large metrics, minimal text. Technical: more detail, still ≤6 objects.
3. Which pillar? Determines color palette.

Step 2: Per-Slide Content Strategy

1. Identify THE one message
2. Find most important element → make it largest (metric? insight? visual?)
3. Count objects — ≤6?
4. Text audit — >50 words? → move to speaker notes
5. Redundancy check — slide text duplicates speaker notes? → choose one

Step 3: Visual Design

Layout options: Two-column | Half-bleed image | Large metric callout (60-72pt + small label) | Icon + text rows | 2×2 or 2×3 grid

Every slide needs a visual element: Chart, icon set, product screenshot, customer photo, or organic shape motif. Oracle icon style: icons in colored circles.

Oracle Anti-Patterns (NEVER)

• Generic blue defaults — use pillar-specific colors, not default blue
• Centered body text — left-align paragraphs and lists
• Repeated layouts — vary slide structures throughout deck
• Accent lines under titles — hallmark of AI-generated slides, violates Oracle brand
• Text-only slides — every slide needs a visual element
• Multiple messages per slide — confuses audience, kills retention
• More than 6 objects — cognitive overload
• Missing footer — Oracle requires footer on all slides
• Wrong background for context — dark for live events, not documents
• Arial when Oracle Sans Tab available — always check font availability first

Template Rules

Accessibility (Oracle Standard)

Reading order (screen readers read bottom to top in Selection Pane):

1. Home → Arrange → Selection Pane
2. Drag most important content to bottom of list

Alt text: Right-click image → "View Alt Text". Describe content for blind/low-vision users. Format:

"Bar chart showing 127% revenue growth YoY"

Contrast: Minimum 4.5:1 (WCAG AA). Oracle palettes pre-selected for strong contrast.

Cognitive Load Validation Checklist

Before finalizing each slide:

• One message — summarizable in one sentence?
• ≤6 objects (images + shapes + text boxes + charts, excluding footer)?
• ≤50 words on slide (verbose content in speaker notes)?
• Largest element = most important content?
• No redundancy between slide text and speaker notes?
• Contrast ≥4.5:1?
• Background matches context (dark live / light docs / mixed virtual)?
• Visual element present?
• Oracle Sans Tab font used?
• Correct pillar color palette?
• Footer present with current year and confidentiality level?
• Alt text on all images?
• Selection Pane reading order correct?

QA Workflow (Required)

Mindset: Assume problems exist. First render is almost never correct.

Phase 1: Content QA

python -m markitdown output.pptx
python -m markitdown output.pptx | grep -iE "xxxx|lorem|ipsum"

Phase 2: Cognitive Load QA

Per slide: object count ≤6? word count ≤50? one clear message? largest = most important?

Phase 3: Brand Compliance QA

Oracle Sans Tab? Correct pillar palette? Background matches context? Footer correct year? No AI patterns (accent lines, centered body text)?

Phase 4: Visual QA with Subagents

# Convert to images for visual inspection
python scripts/office/soffice.py --headless --convert-to pdf output.pptx
pdftoppm -jpeg -r 150 output.pdf slide

Subagent prompt:

Visually inspect these Oracle-branded slides. Assume there are issues — find them.
Check: overlapping elements, text overflow, object count >6, low contrast,
text-only slides, wrong font, missing/incorrect footer, wrong background for context.
Report ALL issues per slide.

Verification Loop

1. Generate → convert to images → inspect
2. List ALL issues (cognitive + brand + visual)
3. Fix issues
4. Re-verify affected slides
5. Repeat until no issues found

Do not declare success until at least one fix-and-verify cycle is complete.

Reading/Inspecting Content

python -m markitdown presentation.pptx        # Text extraction
python scripts/thumbnail.py presentation.pptx  # Visual thumbnail grid
python scripts/office/unpack.py file.pptx dir/ # Raw XML inspection