Pitch Deck

Construct an investor-ready pitch deck from founder inputs. Output is a single self-contained HTML file built on Reveal.js. The narrative is the deliverable; the slide template is the scaffold that keeps the narrative honest.

What this skill produces

Every run produces exactly THREE files inside the resolved pitch root (see Phase 0 Step 0.0 for path resolution; default is docs/pitch/ for solo runs, docs/startup-kit/pitch/ when invoked via orchestrator):

1. deck.html — single self-contained HTML file. Reveal.js-based, keyboard nav (←/→/Space/Esc), AAA contrast for projection, prints cleanly to PDF via ?print-pdf query param + CSS paged media, zero network dependencies (all JS/CSS inline, images base64 if any).
2. speaker-notes.md — per-slide speaker notes with the 15–30 second spoken version of each slide's message and the most likely investor pushback.
3. deck-checklist.md — before-you-send-this-deck checklist covering the required-slot gate, the 90-second read test, and legal/disclosure reminders.

All three paths are named explicitly in the response (see Phase 3 Step 5). Single-slide reworks still produce three files — the untouched files are regenerated from prior content so the pitch folder stays consistent. All three files go into the resolved pitch folder.

What this skill is NOT

• Not a template you fill in later. Slides ship with real content or with explicit [fill in: …] slots that block rendering until answered. This skill owns deck construction end-to-end — brand-workshop does not ship a deck template.
• Not a financial model. The Ask and Use of Funds slides reference numbers the founder provides; this skill does not produce projections, cohort curves, or unit economics beyond what the founder inputs.
• Not a substitute for customer validation. Traction claims are only as good as the founder's data. The skill flags unsubstantiated traction but doesn't go validate it.
• Not a design tool. Visual polish comes from the brand kit. If no brand artifact is present (<brand-root>/DESIGN.md), the deck uses neutral- professional defaults — don't try to invent a visual identity inside this skill.

Skill Boundaries

This skill intentionally overlaps with team-composer (@startup_strategist and @vc_partner are active there too) but differs in deliverable:

• Use pitch-deck when: the founder wants a shippable artifact — an HTML deck they can attach to an investor email or present from a laptop.
• Use team-composer with @startup_strategist + @vc_partner when: the founder wants a discussion on narrative structure, investor positioning, or one slide without committing to a full deck.

Companion plugins:

• brand-workshop — upstream. Produces brand artifacts (under <brand-root>/ per the conventions doc, or legacy brand-kit/) which this skill reads for visual tokens (DESIGN.md), positioning/voice (brand-brief.md), and taglines (descriptions.md). Brand-workshop does not pre-emit a deck template; this skill generates its own Reveal CSS from the DESIGN.md tokens.
• validation-canvas — upstream. If the validation canvas exists (<canvas-root>/validation-canvas.md or legacy validation-canvas.md), this skill reads it directly to seed the Problem, Solution, and Business Model slides and to stress-test the Ask slide against the Stress Tests section.
• riskiest-assumption-test — direct upstream (heavy gate). If the assumption-test plan exists (<rat-root>/assumption-test-plan.md or legacy rat/assumption-test-plan.md), this skill reads its ## Top 3 Hypotheses and ## Results sections to inform the Validation slide and stress-test the Traction claims. The deck refuses to ship as a clean investor draft unless the RAT Results are populated for the top-3 hypotheses (override with [PRE-VALIDATION DRAFT] watermark — see Phase 0 below).

An umbrella startup-launch-kit plugin may later orchestrate all of these.

The Standard 10-Slide Arc

Investors read decks in under 4 minutes. One claim per slide, numbers where numbers belong, faces on the team slide, a specific ask. The standard arc:

Optional slides (add only if the founder has real content): Why Now, Vision, Appendix (cohort data, cap table, legal structure). Adding optional slides without real content weakens the deck — resist the urge to pad.

See references/slide-contracts.md for each slide's required slots, anti-patterns, and the investor-read-test question it must pass.

Phase 0: RAT gate (heavy gate on validation results)

Goal: refuse to ship a clean investor deck built on untested assumptions. A pitch deck without validation is sales theater.

Step 0.0 — Path resolution + manifest awareness (v2.2.0+)

Resolve the pitch root once at invocation, in this precedence order (canonical chain):

1. Explicit output_dir arg (passed by startup-launch-kit) → use as-is.
2. STARTUP_KIT_DOCS_ROOT env var set → ${STARTUP_KIT_DOCS_ROOT}/pitch/.
3. Smart default — docs/startup-kit/ exists → docs/startup-kit/pitch/. Surface the smart-default notice: "Writing to docs/startup-kit/pitch/ (smart default). Set STARTUP_KIT_DOCS_ROOT=./docs to write standalone instead."
4. Solo fallback → docs/pitch/.

The same root resolution applies to sibling reads (<canvas-root>/, <rat-root>/, <brand-root>/) — they're siblings of the pitch root.

Manifest awareness. Look for kit-manifest.json at <resolved-kit-root>/kit-manifest.json first; fall back to the working-directory root for backward compat. Use it as a hint, never as a bypass:

• Override-flag (special case for this skill): if the manifest's gate_overrides[] array contains an entry with gate: "pitch-deck-pre-validation" and founder_acknowledged: true, honor the override silently (proceed with [PRE-VALIDATION DRAFT] watermark per Step 0.2 below) but surface a one-line acknowledgment: "Manifest records a [PRE-VALIDATION DRAFT] override from [date] — reason: [reason]. Proceeding with watermark." Do NOT ask the founder to re-confirm; the manifest entry IS the consent record.
• If the manifest lists pitch-deck as completed with a recent mtime, surface that fact: "Manifest says you ran pitch-deck on [date]. Update mode (single-slide rework or full rebuild), fresh run, or skip to grill?"
• Manifest read failures (corrupt JSON, missing fields) are non-fatal — log the issue inline and proceed as if no manifest exists.

After this skill ships its artifacts (Phase 3 — build & ship), if a manifest exists, append/update this skill's entry. If a new override was used during this run (founder declared a pre-validation draft), record it in gate_overrides[] with timestamp + reason + founder_acknowledged: true. Use atomic write (write .tmp, then rename). If the manifest doesn't exist, do NOT create it — that's the startup-launch-kit orchestrator's job. See startup-launch-kit/references/manifest-schema.md for the schema and startup-launch-kit/references/gate-override-protocol.md for the override format.

Step 0.1 — Check for the assumption-test plan (heavy gate)

Look for assumption-test-plan.md at <rat-root>/assumption-test-plan.md first (sibling of pitch root); fall back to the legacy path rat/assumption-test-plan.md (cwd-relative) for backward compat. If found at the legacy path, surface a one-line notice so the founder knows the artifact is at a v1 location: "Read RAT plan from legacy v1 path."

If the file is missing at both paths, STOP and route the founder to riskiest-assumption-test:

"This skill needs the assumption-test plan with populated ## Results for the top-3 hypotheses. Pitching on untested assumptions is sales theater — investors spot it instantly. Run riskiest-assumption-test first, then come back here when at least one top-3 hypothesis has a confirmed result."

If the file is present but ## Results is empty for all top-3 hypotheses, STOP and route the founder to run the tests they planned.

Step 0.2 — Override path: pre-validation draft

The founder can explicitly override the gate by saying "this is a pre-validation draft", "I'm pitching for early feedback", "I haven't tested yet — show me what the deck would look like", or similar. In that case:

1. Proceed with Phase 1 normally.
2. Add a [PRE-VALIDATION DRAFT] watermark to slide 1 (Title) and slide 0 (the warning slide if any cardinal slots are unfilled).
3. Add a footer line on every slide: "Pre-validation draft — assumption-test-plan Results pending."
4. Output <pitch-root>/deck-checklist.md with a section explicitly listing what's NOT validated, so the founder owns the override decision.

Step 0.3 — Light gate (always, even with full RAT)

Even when RAT Results are populated, scan for invalidated hypotheses:

• If any top-3 hypothesis is invalidated in ## Results AND the canvas has not been updated since (mtime check on <canvas-root>/validation-canvas.md vs. <rat-root>/assumption-test-plan.md, or their legacy fallback paths), surface this and STOP:

"Hypothesis [N] in your RAT was invalidated, but the validation canvas hasn't been updated since. The deck would carry the invalidated belief. Run validation-canvas in update mode first, then come back here."

This is the loop-back protocol enforcement at the deck stage. See validation-canvas/references/folder-contract.md.

Phase 1: Intake

Goal: inventory what's already written and figure out the narrative gaps.

Step 1 — Read the working directory

Check for these files, in order. For each, look at the new conventions path first; fall back to the legacy path for backward compat:

If any sibling is found at the legacy path, surface a one-line notice so the founder knows the artifact is at a v1 location: "Read upstream artifact from legacy v1 path."

What to extract from each:

1. Validation canvas (from validation-canvas) — if present, parse the ### Customer Segments, ### Unique Value Proposition, ### Solution, ### Revenue Streams, ### Customer Jobs, ### Customer Pains, and ## Stress Tests sections. These seed slides 2 (Problem — from ### Problem
   • ### Customer Pains), 3 (Solution — from ### Solution + ### Pain Relievers), 6 (Business Model — from ### Revenue Streams + ### Cost Structure), 7 (Traction — cross-checked against RAT Results below), and the anti-patterns check.
2. Assumption test plan (from riskiest-assumption-test) — if present, read ## Top 3 Hypotheses (each hypothesis seeds an optional "Validation" slide before Traction) and ## Results (the confirmed results become evidence on the Traction slide; invalidated results trigger the loop-back gate from Phase 0).
3. Brand design system (from brand-workshop) — if present, read the YAML front matter at the top of DESIGN.md (see spec: https://github.com/google-labs-code/design.md, version: alpha) and extract color tokens, typography, and logo path for deck styling. Map colors.primary to the deck's --deck-accent CSS variable; map typography.h1.fontFamily to headings (or typography.body-md.fontFamily as fallback).
4. Brand brief (from brand-workshop) — if present, read the ## Positioning and ## Voice & Tone sections (named anchor headings) for tone matching. If those sections are missing, fall back to the ## Executive Summary + ## Workshop Transcript sections.
5. Brand descriptions (from brand-workshop) — if present, read the tagline row and short/medium bios. Slide 1 (Title) uses the tagline from here first, falling back to brand-brief.md → Final Concept → Tagline.

Note on deck CSS: brand-workshop does not emit a deck template or CSS file. This skill owns deck construction end-to-end. Generate the Reveal CSS directly from the tokens in DESIGN.md (color, typography, spacing) via the YAML front matter. Do not look for <brand-root>/deck/pitch-styles.css or <brand-root>/deck/pitch-template.html (or their legacy brand-kit/deck/ equivalents) — those artifacts no longer exist.

If the validation canvas doesn't exist: see Phase 0 — RAT gate already STOPped you. If the assumption-test plan doesn't exist, same. If only the brand artifacts are missing: proceed with neutral defaults; brand assets are optional, validation is not.

Step 2 — Classify the deck variant FIRST

Before any slide drafting, ask exactly ONE question and wait for the answer:

"Which variant: (a) pre-seed, (b) seed, (c) Series A/B, or (d) demo-day / cold-send / warm intro / follow-up? This changes how I treat traction and the ask."

Do not skip. Do not infer from context. Do not proceed to Step 3 without an answer. The variant gates, at minimum:

• Traction slide depth — pre-seed = validation proxies (LOIs, waitlist, pilots with end dates); Series A/B = revenue curve, retention cohorts
• Ask slide format — pre-seed = SAFE/note, round size only; Series A/B = priced round, use-of-funds table, runway in months
• Competition framing — pre-seed = "why now + insight"; Series A/B = "why us beats them, with evidence"
• Length tuning — see references/deck-variants.md

If the user refuses to pick, default to seed and state that assumption verbatim in the first line of the response: "Assuming seed variant — correct me if wrong."

Step 3 — Identify what's missing

List the slides whose required slots (references/slide-contracts.md) aren't yet answered by inputs from the validation canvas, the assumption-test plan, or prior founder content. Show the list to the founder before launching into the interview — this prevents surprise and lets the founder batch context.

Phase 2: Narrative Interview

Goal: fill every required slot with content the founder actually owns.

Role setup

Run in first-person voice with these roles active:

Interview protocol

Go in the standard arc order (Title → Ask), slide by slide. For each slide:

1. Read the slide's required slots from references/slide-contracts.md.
2. Show the founder the required slots. Collect real answers in their words — do not paraphrase into VC-speak.
3. Every 2–3 slides, run the @vc_partner read test. They read what's been produced so far as if it were cold, and flag what doesn't land. If something doesn't land, revise before continuing — not at the end.
4. Surface the top anti-pattern for the slide (see references/slide-contracts.md). Explicitly check the deck doesn't fall into it.

What to do when the founder doesn't have an answer: do not invent. Mark the slot [fill in: ${specific-question}]. The deck-checklist.md will block shipping until these are filled. See "Required-Slot Gating" below.

The four cardinal sins (flag and refuse)

The skill must refuse to ship a deck that commits any of these:

1. TAM-only market sizing. A market slide with only TAM (top-down) and no bottom-up SAM or SOM is a lie by omission. Refuse to render without both.
2. Traction without a time axis. "10,000 users" could be over 10 years or 10 days. Every traction number has a start date.
3. Team slide with no faces or no relevant experience. Faces + prior-experience tag per founder. Missing either is a flag.
4. Vague ask. "Raising $X to grow the business" fails. Required: amount, milestones the amount buys, runway in months.

These are hard gates. If the founder wants to ship anyway, document the override in deck-checklist.md so they own the decision.

Single-slide rework mode

When the user asks to rework ONE slide without rebuilding the full deck (e.g., "just help me with my Competition slide", "rework my Market slide"):

1. Apply that slide's contract from references/slide-contracts.md as usual — including its required visual format (e.g., Competition needs a 2×2 or comparison matrix, not a bullet list; Traction needs a chart with a time axis, not a number-in-prose). Name the format requirement in your first response so the founder knows what they're ultimately producing, even while you're still gathering intake content.

2. Cross-check against the named adjacent slides before finalizing. An isolated rework is how a single-slide edit quietly breaks the rest of the deck. Standard adjacencies:

   Slide being reworked	Must stay consistent with
   Problem	Market (SAM named segment), Solution (fits this pain)
   Solution	Problem (addresses stated pain), Product (matches demo)
   Market	Problem (same segment), Business Model (ACV × SOM math)
   Product	Solution (what the demo proves), Business Model (the action pricing hangs on)
   Business Model	Traction (metric must match the model), Market (SOM matches ACV × customer count)
   Traction	Business Model (MRR vs GMV vs users — match the model), Ask (trajectory supports runway)
   Team	Ask (team has skills to execute the milestones)
   Competition	Problem (segment), Solution (differentiation axis), Product (claim is demonstrable)
   Ask	Traction (trajectory supports milestones), Team (can execute), validation canvas Stress Tests, assumption-test plan Results

3. Ask the founder briefly for the adjacent slides' current content, or read them from prior drafts if provided. Do not rework in isolation.
4. The four cardinal sins still apply to the single slide being reworked — e.g., a standalone Market rework is still gated on SAM + SOM bottom-up; a standalone Traction rework is still gated on a time axis.

Phase 3: Build & Ship

Goal: produce <pitch-root>/deck.html, <pitch-root>/speaker-notes.md, and <pitch-root>/deck-checklist.md. The <pitch-root> is resolved per Phase 0 Step 0.0. Create the folder if absent.

Step 1 — Assemble the Reveal.js HTML

Follow the template pattern in references/deck-template.md. Key constraints:

• Single file. Reveal.js core, theme CSS, plugin JS, and any images all inline or base64. No <script src="https…">. No <link rel="stylesheet" href="https…">.
• Brand token substitution. If the brand design system is present (<brand-root>/DESIGN.md), read the YAML front matter and substitute color and font tokens into the Reveal theme via CSS custom properties. Map colors.primary → --deck-accent, typography.h1.fontFamily → headings. Check contrast for projection (AAA for body text, AA minimum for accents).
• Print-to-PDF. Include @media print rules that hide controls, force one slide per page, and respect page breaks. Test that appending ?print-pdf to the file URL produces a clean slide-per-page PDF when the user prints from the browser.
• Charts. Use inline SVG from references/deck-charts.md. No external chart library. Market bar/stack, Traction line with time axis, Competition 2×2 — these three cover ~95% of deck chart needs.
• Images. Base64-encode any images the founder provides. If the founder provides image paths instead, copy the files into the resolved pitch folder and reference them relatively — the deck still works offline even if the encoded version isn't used.

Step 2 — Required-slot gating

Before writing the file, scan for any [fill in: …] markers in the content. For each:

• If the slot is a cardinal required slot (see references/slide-contracts.md), render a warning slide at position 0 titled "This deck is not ready to send" listing each unfilled cardinal slot.
• If the slot is non-cardinal (e.g., optional metric, secondary logo), render the slot inline with muted styling and [fill in: …] prompt text.

Never silently ship a hollow deck. Investors spot them instantly and it costs the founder credibility with that firm.

Step 3 — Write speaker notes

For every slide, write <pitch-root>/speaker-notes.md with:

• The 15–30 second spoken version of the slide's message (investor meetings run fast — if a slide can't be delivered in 30 seconds, it's too dense).
• The most likely investor pushback and the one-line response.
• Optional: the deep-dive redirect — what slide or data the founder can pull up if pushback goes deeper.

See references/speaker-notes.md for the format.

Step 4 — Write the deck checklist

Produce <pitch-root>/deck-checklist.md covering:

• Before you send: required slots filled, traction has a time axis, ask is specific, team slide has faces.
• The 90-second read test: the founder reads the deck cold in 90 seconds and verifies they understand the business. If they don't, the deck isn't done.
• Legal / disclosure reminders: "Confidential" in the footer if under NDA, forward-looking-statement disclaimer if projections are shown, CCPA/GDPR if personal data is referenced anywhere.

See references/deck-checklist.md for the template.

Step 5 — Present to the user

Use present_files if available. Otherwise emit clickable computer:// links.

Always name all three output files explicitly in the response, with their full resolved paths:

• <pitch-root>/deck.html (e.g., docs/startup-kit/pitch/deck.html or docs/pitch/deck.html)
• <pitch-root>/speaker-notes.md
• <pitch-root>/deck-checklist.md

Name them even on partial runs — a single-slide rework still names the file being edited. Do not substitute "the deck file", "your slides", or "the notes" for the path.

End with EXACTLY these three lines, in this order, prefixed verbatim:

1. Weakest slide for investor read-test: <slide name> — <why>
2. Strongest remaining anti-pattern: <pattern> on <slide name>
3. Highest-value next edit: <one sentence>

This applies to first-pass decks, single-slide reworks, and regenerations. Do not replace the triad with presentation-prep tips, "next steps", or a summary block. If a line is genuinely N/A (e.g., deck passed every gate), write the prefix + none — never omit the line.

Output Files

<pitch-root>/
├── deck.html              Self-contained Reveal.js deck (primary deliverable)
├── speaker-notes.md       Per-slide 15–30 sec spoken version + likely pushback
└── deck-checklist.md      Before-you-send gate + 90-second read test + legal

Where <pitch-root> resolves per Phase 0 Step 0.0:

• docs/startup-kit/pitch/ — orchestrated (via startup-launch-kit)
• docs/pitch/ — solo default
• docs/startup-kit/pitch/ — solo with docs/startup-kit/ smart default
• ${STARTUP_KIT_DOCS_ROOT}/pitch/ — env-var override

No other files. Do not scatter intermediate HTML drafts.

Quality Checklist

Before presenting to the user, verify each:

Narrative

• Every slide answers the core question from the standard arc (one claim per slide)
• Market slide has both bottom-up SAM and SOM, not just TAM
• Traction slide has a time axis on every number
• Team slide has faces + relevant prior experience per founder
• Ask slide names amount, milestones, and runway in months
• No [fill in: …] cardinal slots remain — or a warning slide at position 0 lists them

Rendering

• deck.html is a single file — no external CSS/JS/font/image URLs
• Keyboard navigation works (←/→/Space/Esc)
• ?print-pdf produces a clean slide-per-page PDF
• AAA contrast for body text against projection backgrounds; AA minimum for accents
• Brand tokens applied if <brand-root>/DESIGN.md present; neutral defaults otherwise

Shipping

• Three files present: deck.html, speaker-notes.md, deck-checklist.md
• Speaker notes have 15–30 sec spoken version + likely pushback per slide
• Checklist has the 90-second read test and legal reminders
• Files saved to the resolved pitch folder per Phase 0 Step 0.0 (not cwd root)
• Smart-default notice surfaced if smart-default fired
• Response ends with: weakest slide + worst remaining anti-pattern + single highest-value next edit

Cross-Skill Integration

Principle: this skill owns the investor-facing HTML deck. It does not do financial models, does not do customer validation, does not do brand identity. Upstream plugins feed it; downstream decisions (negotiation, term sheets) are @vc_partner territory in team-composer.

Graceful degradation: if a referenced skill is not installed, this skill still ships a complete pitch folder — downstream integrations are enhancements, not requirements.