pbi-to-exec-deck: Instructions file for Claude Code

Claude Project Instructions

Project Overview

Converts Power BI dashboard exports (.pptx, .pdf, .pbip) into executive-ready 16:9 PPTX presentations using Claude's vision and analytical capabilities. No API key needed — uses current Claude Code session.

Workflow

python convert_dashboard.py "dashboard.pptx"             # Full run → dashboard_executive.pptx
python convert_dashboard.py "file.pptx" --prepare        # Step 1 only (extract)
python convert_dashboard.py --build --output "exec.pptx" # Step 3 only (build)
python convert_dashboard.py --verify                     # Validate insights.json

Three steps:

1. Extract (auto, ~5s): Images → temp/slide_N.png, metadata → temp/analysis_request.json. PPTX skips first slide (title/metadata); PDF includes all pages.
2. Analyze (you): Read images → generate insights → write temp/insights.json.
3. Build (auto, ~3s): Renders slides, validates against Constitution.

Rendering modes:

Mode	Default	Flag	Output
Screenshot	✅	—	PBI screenshot embedded + insight commentary
Vector charts	—	--vector-charts	matplotlib charts from chart specs in insights.json

Use --vector-charts when: exact DAX data available, screenshots are low quality, or specific data points need highlighting. Keep default when: PBI visuals are clean, or complex chart types (dual-axis, heatmaps) are hard to reconstruct.

---

Analysis Task

1. Read temp/analysis_request.json
2. Read temp/slide_N.png for each slide listed
3. Act as senior analyst advisor to an IT decision maker
4. Write temp/insights.json

CRITICAL: Read docs/DASHBOARD_READING_RULES.md before analyzing.

insights.json Root Structure

{
  "deck_title": "Compelling title: specific finding or story",
  "deck_subtitle": "Platform · Org · Month Year – Month Year",
  "executive_summary": [
    "Finding with specific number → business implication"
  ],
  "recommendations": ["Action verb + what to do + expected outcome"],
  "slides": [ /* per-slide objects */ ]
}