Deck Pipeline

A 4-agent pipeline that transforms a topic into a critique-proof presentation deck. Each agent has a distinct role — the output of one feeds the next. The result is a deck that has been strategically structured, built, reviewed, and fixed before you ever touch it.

The Four Agents

Topic
  │
  ▼
┌─────────────┐    ┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│ STRATEGIST  │───▶│   BUILDER   │───▶│   CRITIC    │───▶│    FIXER    │
│             │    │             │    │             │    │             │
│ Defines the │    │ Creates the │    │ Reviews like│    │ Applies the │
│ narrative   │    │ actual deck │    │ a McKinsey  │    │ top 3 fixes │
│ and outline │    │             │    │ EM at 2am   │    │             │
└─────────────┘    └─────────────┘    └─────────────┘    └─────────────┘

Agent 1: Strategist

Role: Define the audience, governing thought, and full slide outline before anything gets built.

What it does:

• Identifies the target audience and what they care about
• Defines the governing thought (the one sentence the whole deck proves)
• Creates a 10-12 slide outline using Situation → Complication → Resolution structure
• Each slide title is a claim, not a label ("RTD market will reach $42.5B by 2028" not "Market Overview")
• Groups slides into clear narrative blocks

Output: A structured outline with:

• Audience definition
• Governing thought
• Slide-by-slide titles with supporting points
• Narrative flow markers (Situation / Complication / Resolution)

Why this matters: A deck without a strategy is just a collection of slides. The strategist ensures every slide earns its place in the argument.

Agent 2: Builder

Role: Convert the strategist's outline into an actual presentation.

What it does:

• Takes the outline and creates the full deck content
• Writes slide body text, data points, and supporting evidence
• Structures visual elements: comparison tables, stat callout boxes, framework diagrams
• Maintains consistent formatting and tone throughout
• Ensures each slide has a clear "so what"

Output: A complete deck as a native .pptx (via python-pptx) with editable text and editable charts. Optional: markdown or formatted text as an intermediate artifact for review.

• Claim-based titles on every slide
• Data points with sources where available
• Visual structure suggestions (tables, callout boxes, charts)
• Section transitions between narrative blocks

Tools the Builder uses:

• mckinsey-charts for native editable charts — bar+callout for single-number stories, stacked column for mix over time, waterfall for bridges. The audience can right-click → Edit Data inside PowerPoint. Use this instead of pasting chart images.
• python-pptx for native text, tables, and layout. No screenshots, no image-based content — everything stays editable.

Tips for the Builder:

• Stat callout boxes work well for key numbers (large font, contrasting color)
• Comparison tables > bullet lists when presenting options
• Keep body text to 3-5 points per slide maximum
• Every data slide needs at least one source
• Every quantitative slide uses a chart from mckinsey-charts, not a screenshot

Agent 3: Critic

Role: Review the deck like a senior engagement manager would — find every structural problem, weak argument, and missing element.

What it does:

• Grades the overall deck (Green / Yellow / Red)
• Reviews each slide with a status and specific issue
• Identifies the top 3 fixes ranked by impact
• Calls out one thing that works (to protect it)

Grading:

What it checks:

• Narrative flow and structure
• Data rigor (sources, specificity, comparisons)
• "So what" on every slide
• Financial framing where needed
• Claim-based titles vs topic titles
• Frameworks vs bullet lists

Output: A critique document with section-by-section grades and top 3 prioritized fixes.

See the McKinsey Critic skill for the full review framework.

Agent 4: Fixer

Role: Apply the critic's top 3 fixes to produce the final deck.

What it does:

• Takes the critic's specific fix instructions
• Applies each fix to the deck
• Documents what changed and why
• Produces the final, polished version

Common fixes the Fixer applies:

• Adding source footnotes to data slides
• Rebuilding bullet lists as comparison tables (columns = options, rows = evaluation dimensions)
• Adding financial frames (market size × share = revenue, margin analysis, investment required)
• Sharpening headlines from descriptions to claims
• Naming competitors instead of describing them generically
• Linking disconnected slides back to the core argument

Output: The final deck with all fixes applied, plus a change log showing before → after for each fix.

How to Use

Quick Start

"Build me a deck on [topic] for [audience]. Use the full pipeline — 
strategist, builder, critic, fixer."

With More Context

"I need a deck for [audience] arguing that [governing thought].

Key data points I have: [list any data, stats, or sources you want included]

Run the full pipeline: outline it, build it, critique it, fix it."

Step by Step (if you want control between stages)

Step 1: "Act as the strategist. Define the audience, governing thought, 
        and 10-slide outline for [topic]."

Step 2: "Now act as the builder. Turn this outline into a full deck."

Step 3: "Now act as the McKinsey critic. Grade this deck and give me 
        the top 3 fixes."

Step 4: "Now act as the fixer. Apply the top 3 fixes."

Example: What Each Agent Produces

Topic: "Should a premium coffee company enter the RTD (Ready-to-Drink) market?"

Strategist output:

• Audience: Board of directors + CEO
• Governing thought: "The premium RTD market offers a $425M revenue opportunity that plays to our existing capabilities — but only if we move within 12 months."
• 10 slides: 3 Situation (market size, specialty growth, RTD growth) → 3 Complication (consumer shift, competition, first-mover window) → 4 Resolution (path comparison, RTD recommendation, sustainability, roadmap)

Builder output: Full deck with data, stat callouts, comparison tables

Critic output:

• Grade: YELLOW
• Fix 1: Cite every number (no source footnotes on any slide)
• Fix 2: Slide 7 is a bullet list — needs comparison table (3 paths × 5 dimensions)
• Fix 3: No financial frame — add $425M revenue at 1% share, $15-25M Phase 1 investment

Fixer output: Final deck with all sources added, comparison table rebuilt, financial frame on recommendation slide

When to Use

• Board presentations and investor updates
• Strategy recommendations to leadership
• Product launch proposals
• Market entry analysis
• Any deck that needs to survive tough questions

What replaced the old Gamma path

Earlier versions of this repo paired the storyline with the Gamma API to generate finished decks. The current version uses native python-pptx + the mckinsey-charts skill instead. Reasons:

• Editable, not generated. Charts are real PowerPoint chart objects, not images. The audience can change colors, edit data, copy to their own deck.
• No third-party dependency. No Gamma account, no API key, no per-deck credit cost.
• Better defaults. Single highlight color, no gridlines, claim-shaped titles — already wired into mckinsey-charts. The Gamma path required prompt-engineering the look every time.

The gamma-automation/ folder is preserved for reference but is deprecated.

Why a Pipeline Instead of One Prompt

A single "build me a deck" prompt produces mediocre work because it tries to think strategically AND write content AND self-critique simultaneously. By separating the roles:

• The strategist focuses purely on narrative architecture
• The builder focuses on content and visuals, with a clear brief
• The critic is adversarial — it's not trying to protect what was built
• The fixer has specific, prioritized instructions — not vague "make it better"

The result is a deck that has been through the equivalent of a junior consultant → senior associate → engagement manager review cycle.