presentation-factory-humanize

name	presentation-factory-humanize
description	Presentation-Factory-specific humanize wrapper over consulting:consulting-humanize. Strips AI tells and consultant jargon, plus adds slide-specific patterns: title brevity, stat-callout register, eyebrow rules, speaker-note voice, and body-prose density limits. Mandatory final pass before any deck is composed, and again after Compose before the deck ships. Operates on 10_Content.md (Draft stage) and on rendered slide HTML (post-Compose).
type	skill
plugin	presentation-factory
tags	["type/skill","plugin/presentation-factory","scope/specialty","topic/voice"]
status	active

Goal

Make every deck read like a human wrote it. Strip AI tells. Strip consultant jargon. Enforce slide-format register rules. Two invocations per deck: once after Draft (on 10_Content.md), once after Compose (on rendered slide HTML).

When to invoke

• End of Draft stage (mandatory). Operates on 10_Content.md. Run before Compose generates assets.
• End of Compose stage (mandatory). Operates on rendered slide HTML in slides/*.html. Catches anything the composer introduced.
• Any time the user says "this sounds AI" or "this sounds consultant-y" about a specific slide or section.

Required context

consulting:consulting-humanize (the underlying skill this wraps) and references/presentation-factory-foundations.md (typography rules for the slide-format-specific patterns).

Two passes

Pass 1: General humanize (delegate to consulting:consulting-humanize)

Apply all the rules from consulting:consulting-humanize:

• Remove em-dashes and en-dashes (use periods, commas, parens, or "to" for ranges)
• Break sentences over 25 words
• Strip filler verbs: harness, navigate, unlock, empower, leverage, utilize, facilitate
• Strip filler adjectives: robust, seamless, comprehensive, cutting-edge, transformative
• Strip "not X but Y" comparative pattern
• Strip "This..." summary closers
• Strip throat-clearing openers
• Strip hedging stacks
• Active voice over passive
• Plain words over Latinate

Don't re-explain the rules. Cite the source skill.

Pass 2: Slide-format-specific patterns

These rules are layered on top of the general humanize pass.

Title brevity (slide H1)

Length	Verdict
1 to 5 words	Ideal for presentational decks
6 to 9 words	Acceptable, ideal for informational decks
10 to 14 words	Borderline. Likely too dense; try to cut.
15+ words	Too long. Break into title + subtitle.

Slide titles must have exactly one <em> emphasis phrase (the signature Slalom move). Zero <em> = brand violation. Two or more = visual chaos.

Subtitle register

Subtitles are one sentence. Avenir Next bold sentence case. They support the title; they don't restate it. If the subtitle is a longer version of the title, rewrite the subtitle.

Eyebrow rules

Eyebrows are UPPERCASE, 1 to 4 words. Letter-spacing handled in CSS. Always brand-primary color. Examples that pass: BOTTLENECK #1, CUSTOMER STORY, SECTION 2 OF 3. Examples that fail: Important Information, Welcome to our presentation, the speed tax (must be UPPERCASE).

Stat-callout voice

Big stats are bare numbers with units. Below the number, a single phrase frames it. No "approximately," no "over X%," no soft language.

Bad: "Approximately 90% of AI pilots fail to scale, according to McKinsey research." Good: "90% of AI pilots fail to scale." (Source attribution as a separate small line.)

Body prose density

In presentational decks: any single body block over 30 words gets cut. In informational decks: any single body block over 60 words gets cut. Bullet lists: max 5 items. If you need 6+, split into two slides or merge concepts.

Pull quote attribution

Format: *"Quote text."* First-name Last-name, Role, Organization. No em-dash before the attribution. No "said." If the role is too long, drop the role; keep name + organization.

Speaker note voice

Speaker notes are 3 to 5 sentences. Conversational, first-person from the speaker's perspective. They should sound like someone speaking, not someone reading. Strip:

• "In this slide we'll see..."
• "I'd like to take a moment to discuss..."
• "As you can see..."

Replace with direct speech: "The number that matters here is 7 billion. Here's why."

Footer chrome

Don't humanize footer chrome (slide numbers, deck title, brand lockup). Those are structural, not prose.

Workflow

1. Read the input (10_Content.md for Pass 1 invocation, rendered HTML for Pass 2)
2. Apply Pass 1 (general humanize rules from consulting:consulting-humanize)
3. Apply Pass 2 (slide-format-specific patterns)
4. Write changes in place. Do not create a "humanized" copy. The original file gets updated.
5. Log a brief report at the end:
   • Em-dashes removed: N
   • Filler verbs removed: list
   • Titles flagged for length: list of slide numbers
   • Body blocks over density limit: list of slide numbers
   • Eyebrows fixed: list of slide numbers
6. Save log to WIP/[deck]/humanize.log

Anti-patterns

• Don't change meaning. If a sentence carries real substance, keep the substance and rewrite the form.
• Don't strip the brand's signature italic-emphasis pattern. *Title <em>emphasis</em>* is a styling rule, not an AI tell.
• Don't humanize technical structural markers (template names, eyebrows in caps, data slots in HTML).
• Don't add throat-clearing of your own. ("I've reviewed the deck and noticed that..." — no.)
• Don't humanize the same file twice in one session without prompting. The cleaned file is the new baseline.

Output

Updated 10_Content.md (after Pass 1) or updated slides/*.html files (after Pass 2). Plus WIP/[deck]/humanize.log summarizing changes.