presentation-factory-draft

name	presentation-factory-draft
description	Stage 2 orchestrator for the Presentation Factory — Draft. Calls presentation-factory-content-author to write the slide-by-slide content; presentation-factory-speaker-notes for notes; consulting + marketing + brand + humanize + design-taste skills for content quality. Pauses for user review at the end. Output: 10_Content.md.
type	skill
plugin	presentation-factory
tags	["type/skill","plugin/presentation-factory","scope/orchestrator","topic/draft"]
status	active

Goal

Produce WIP/[deck]/10_Content.md — every slide's content + asset prompts + speaker notes, validated against brand voice and stripped of AI tells.

When to invoke

Called from presentation-factory-create-flow after Frame is complete, or directly via /presentation-draft.

Pre-flight

Verify:

• WIP/[deck]/00_Intake_Brief.md exists and has all required sections
• WIP/[deck]/brand/tokens.css and logo.svg exist

If anything's missing, route back to Frame and stop.

Steps

Step 1 — Hand off to presentation-factory-content-author

Invoke the presentation-factory-content-author skill. It walks its own 5-stage process:

1. Outline the deck arc
2. Map slides to templates
3. Draft each slide's content with italic-emphasis applied
4. Specify image / infographic / icon prompts
5. Run consulting + humanize + brand-voice + taste-check passes

The skill returns when 10_Content.md is drafted (without speaker notes if user opted to defer them).

Step 2 — Apply speaker notes

Invoke presentation-factory-speaker-notes to ensure every slide block has a **Speaker notes**: section. This skill either drafts notes that are missing or refines weak existing notes.

Step 3 — Final humanize pass on the full deck

Invoke consulting:consulting-humanize once more against the complete 10_Content.md file. This catches any AI-tell phrases that slipped through per-slide passes (looking at headlines holistically across the deck — e.g. "elevate" appearing in three different slide titles is something you only catch reading them together).

Step 4 — Taste-check the deck

Invoke design:design-taste against the deck-level question: does this deck have variance? If 60%+ of slides use the same template, push back via user prompt to redistribute. If every title uses italic-emphasis on the same word ("impact" 8 times), suggest alternatives.

Step 5 — Generate the deck-table-of-contents

At the top of 10_Content.md, write a simple TOC:

# Deck: [name]
**Style**: [informational | presentational]
**Brand**: [slalom | composable-dxp | client:NAME]
**Audience**: [from intake]
**Runtime**: [from intake]
**Slide count**: [actual]

## Table of contents
1. Title — Build *better tomorrows* for ACME
2. Agenda — What we'll *cover*
3. Section: The problem
4. The market is *shifting* faster
...

Step 6 — Surface for review

Return path to 10_Content.md for user review. Tell the user:

• Total slide count
• Approx. asset count (images + infographics + icons that need to be generated)
• Estimated compose time

Pause via the main orchestrator's pause point.

Output

WIP/[deck]/10_Content.md — slide-by-slide content, asset prompts, speaker notes, deck metadata header.

Anti-patterns

• Don't skip the deck-level humanize pass — per-slide passes miss cross-slide patterns
• Don't generate assets at this stage — that happens in Compose. Content first.
• Don't decide the slide count before the arc — arc decides count
• Don't write content that doesn't map to a specific template — if a slide doesn't fit any of the 25, the content has the wrong shape (split it or merge it)