pitch-deck - SKILL.md Agent Skill

name: pitch-deck description: Governs structure, headlines, claims, and persuasion for ANY pitch deck or investor-facing slide narrative — apply whenever creating, writing, reviewing, critiquing, restructuring, tightening, or rendering a pitch deck. ALWAYS invoke before drafting or revising an investor deck, fundraising deck, seed/pre-seed deck, accelerator or YC / a16z Speedrun application deck, demo-day deck, a deck's headlines / spine / cover / closing slide, or any slide-based document meant to persuade an investor, partner, or selection committee. Trigger phrases include "pitch deck", "investor deck", "slide deck", "fundraising deck", "deck headlines", "deck review", "rework the slide", "tighten the deck", "YC deck", "a16z deck", "Speedrun deck", "demo day", "cover slide", "team slide", "the ask", "render the deck". Do NOT invoke for internal strategy memos, research briefs, or non-persuasion slide docs (training, status decks) UNLESS they will be shown to investors.

Pitch Deck

Principles for any deck meant to persuade an investor or selection committee. The reader is busy, skeptical, and pattern-matching fast — they scan headlines, not paragraphs, and they probe claims in diligence.

The governing instinct: make every headline a claim that stands alone, and make every claim one you can defend. Punch and honesty are the two axes; most failures break one of them.

When reviewing a deck, run the headline-only test and the claim audit (end of file) first — they catch most problems before line edits.

1. The headline carries the argument; the body only proves it

This is the cardinal rule. The single most leveraged edit on any slide is its headline.

One idea per headline — the slide's single takeaway. If the headline says three things, it says nothing. (Kill run-ons like "Two industries, one workflow — one running our product, one paying.")
It must stand alone. A reader who forgot the previous slide must still get it. A headline that's a follow-on sentence fails: "Tokenrip is that substrate" → "what substrate?" Name the thing.
It must be a claim, not an observation. "Two very different verticals. One substrate." is an observation — there's no assertion to agree or disagree with. Give it a verb and a stake: "If it works for a lender and a hospital, it works anywhere."
State the takeaway; don't tease it. A clever-but-cryptic headline makes the investor think "huh?" — that's a lost slide, not intrigue. Summarize the point in a full sentence.
Two-clause headlines are fine (set-up → payoff), the second clause in the accent color. Use a memorable cadence as house style.
Write headline-first. Nail the headline (the idea), and the body usually falls out of it. If the body won't follow, the headline isn't right yet.

The headline-only test: read just the headlines, top to bottom. They should tell the entire story — demand → blocker → what we built → proof → moat → team → close. If the spine doesn't land from headlines alone, restructure before polishing bodies.

2. Lead with the problem; front-load the pain

Open the problem slide with the stakes — the stuck money, the cost, the pain — before any process diagram or mechanism. Never bury the number at the bottom under a list. The reader should feel why this matters in the first line.

3. Bodies are for scanning, not reading

Partners scan. Cut every sentence that doesn't land the pain, prove the claim, or state the moat.

≤ 3 bullets per content slide. Each a fragment, ≤ ~12 words. Bold the key term. At most one italic punch-line per slide (a deliberate parallel pair can be the exception).
Tighten ruthlessly. Before: "Built with an industry veteran who brings the domain expertise and the rolodex; a deal is sourced outside, modeled, then passed between broker, dealer, borrower, lender, and committee…" → After: "Built with a 20-yr industry partner. A deal touches 5+ parties; its history scatters across 3 silos."

4. Categories, not logos

Don't name competitors by company. Use categories ("extraction tools," "single-vertical platforms," "general-purpose AI"). Naming small competitors makes you look small and invites "why not them?" rabbit holes. Keep altitude. (Keep the real names in a Q&A / spine doc for rehearsal, never on a slide.)

5. Find the broadest frame that's actually true

Don't over-qualify the category. A narrowing label (e.g. "regulated industries") can both shrink the prize and misdescribe one of your own examples — a sharp reader catches the wobble. Pick the broadest binding thread that's genuinely true across all your cases (e.g. "accountability" or "high-stakes decisions" instead of "regulated").
Demote the narrow reason to a supporting line in the body, never the frame.

6. Earn a non-obvious insight

The best decks extract a contrarian, defensible insight from a legible fact ("coding has the best AI product-market fit — because code already had the substrate, not because models love code"). Ask: what here is surprising or contrarian? Reframe the category from a crowded race into an open, defensible position. State the insight as the explanation, the prize as the payoff — both on the slide.

7. Claim discipline — survive diligence

Punch must never outrun truth. A diligence call will pull every thread.

Claim position + insight + velocity — never a finished moat you don't have. If the defensible thing hasn't shipped, say you're building toward it.
Distinguish live / in-build / paying / pre-paying, and never blur them. When the thing that's live isn't the thing that's paying, own the split rather than implying "a live paying product."
Don't borrow others' customers or logos as your own. Your customer's customer (the famous hospital, the marquee brand) is their proof, not yours. Cite it as the pain in the space, not "our customer."
No overclaiming plurals. One paid engagement is "our first paid engagement," not "first paid builds."
A punchier headline must not silently upgrade a claim. If the bold line could read as "deployed in both" when one is in-build, the body must hold the honest split.
Keep conflict-sensitive parties anonymous (a partner who's a current employee of a named firm; an unsigned pipeline). Use a role, not a name.
Verify every number before it reaches a slide. If a stat (a tool count, a revenue figure) isn't sourced, omit it until confirmed. Reconcile inconsistent figures (e.g. "300+" vs "400+") to one.

8. Know what THIS investor buys

Tune the deck to the reader, not a generic template.

YC-type: concrete first product + revenue + demand pull; distrusts "operating-system-for-X" manifestos without a shipped thing. Lead with what's real.
a16z Speedrun-type / accelerators: weight insight + founder + velocity over TAM spreadsheets ("capital is a commodity"). Reward a category-defining swing.
Tune the ask to what the program provides (warm intros, talent, network) — not just money.
Omit what the venue doesn't weight (a bottoms-up TAM slide at pre-seed reads as theater) and say so internally so it isn't mistaken for a gap.
Never quote the fund's own thesis back to them — it's the most echoed move in the pile and reads as a status-loser.

9. The team slide

At pre-seed/accelerator, the founders are the bet — this slide matters more than its position suggests.

It must stand alone. Don't assume the application's separate "experience" fields are read; the slide can't depend on them. (Redundancy with those fields is fine; dependency is the risk.)
Lead each founder with the highest-wattage, venture-legible tokens — exits/acquisitions (with ticker), dollar outcomes, scale numbers, "built X solo in N weeks." Compressed highlight reel, not a copy-paste résumé.
Map a credential to a deck claim where you can ("300+ enterprise sales conversations" → "finds first customers for a category that doesn't exist yet").
If domain experts/partners de-risk the build, frame them as an unfair-access advantage, not headcount.

10. Structure and the close

Beat order: hook → what it is → proof → moat → team → close. When the moat is a future state, proof is its own beat — present-tense evidence carries the most weight, so don't fold it into the moat slide.
Maximally different proof points prove generalization. Two unrelated verticals on one substrate is far stronger than one (one example reads as a point solution).
Cover slide = wordmark + tagline only. No subline, no founder credit, no date clutter. The tagline does the work.
Close on vision, not paperwork. A mic-drop that bookends the tagline beats an ask slide. Pull a demo link / "show-me" artifact onto the close — at the application stage it's worth more than an ask. Compress any ask to one line.

11. Working method

Keep prior versions for side-by-side comparison — never destroy an alternative framing; you'll want to choose between them.
Markdown is the source of truth; render the PDF/HTML from it and keep them in sync. Reuse an existing deck's CSS rather than restyling.
Maintain a companion "spine" doc next to the deck: the thesis, claim discipline, pre-answered objections, what the deck deliberately omits, open risks to pressure-test, sourced proof table, and form-field copy. The deck persuades; the spine defends.
Run all prose through /humanizer before rendering (external-facing copy).
The deck is downstream of the real work. Don't let deck polish displace landing the actual proof (the sale, the signed pilot, the live artifact). One real proof point beats any wording change.

Reusable devices

Two-clause accent headline · "What it is / what it isn't" (✓ / ✕) · 2×2 positioning matrix (only with honest axes — never force it) · honest status callout line · velocity stat row · structural analogy mapping (X → Y) shown as a real correspondence, not a costume.

Quick checklist (run before calling a deck done)

Headline-only test — headlines alone tell the whole story; each is a stand-alone claim, none a "huh?" or a "that what?".
Body rule — ≤3 bullets/slide, ≤~12 words, bold key term.
No competitor named by company.
Claim audit — position-not-finished-moat; live vs in-build vs paying kept distinct; no borrowed customers; no unsourced numbers; conflict-sensitive parties anonymous; punchy headlines don't upgrade claims.
Audience fit — leads with what this investor buys; ask tuned to the venue.
Cover = wordmark + tagline only. Close = vision + demo link.
/humanizer pass, then render from the markdown source.