| name | caveman-explainer-deck |
| description | Turn a single concept into a TikTok-style "caveman explains it" slide deck — one idea per slide, 3 to 5 words of big bold text, clean flat illustrations on a pure white background, built with the Substrate MCP. Use when the user wants to make a TikTok/Reels/Shorts explainer, a "caveman voice" or "explain like I'm dumb" deck, a punchy slide-per-beat video script, or says "make a deck about X" for short-form video. The creator narrates in caveman voice over the slides, so slides carry minimal text and one idea each. |
| metadata | {"version":"1.0.0"} |
Caveman Explainer Deck
Build a short, punchy slide deck that explains ONE concept the way a caveman would — for a TikTok/Reels/Shorts video where the creator narrates in caveman voice and the slides are the visuals behind them.
The whole point: the slide is not the explanation — the creator's voiceover is. So every slide is one idea, a few big words, and one clean illustration. The slides set up beats; the human lands them.
[!important] You need the Substrate MCP
This skill renders slides with the mcp__substrate__* tools. If they aren't connected, tell the user to add the Substrate MCP server first. If the tools are deferred, load them with ToolSearch (see references/substrate-workflow.md).
The only input you need
A concept. That's it. Example invocations:
- "Make a caveman explainer about reward hacking."
- "Caveman deck: how vaccines work."
- "Explain compound interest like a caveman, TikTok format."
Optional extras the user MAY give (use sensible defaults if not):
| Input | Default |
|---|
| Aspect ratio | 16:9 (creator shows a laptop on camera). Use 1:1 if they say it fills a phone screen. There is no 9:16 in Substrate. |
| Slide count | 6 (range 5–8). Tighter is better for retention. |
| Tone / accent | Caveman grunt narration; on-slide text is plain, punchy, ALL CAPS. |
Do not interrogate the user. Take the concept, make the deck, then show renders and offer tweaks.
The procedure
Follow this end to end. Full tool-by-tool detail (arguments, timing, gotchas) is in references/substrate-workflow.md.
- Outline the story spine first (before any tool call). Break the concept into 5–8 beats using the arc in
references/slide-formula.md. Each beat = one slide = one idea = 3–5 on-screen words. Write the on-slide text for every slide now.
create_deck with the title and the chosen aspect_ratio. ⚠️ Aspect ratio is locked at creation — pick correctly the first time; changing it later means a brand-new deck.
set_design_prompt with the exact design system from references/design-system.md (paste it verbatim — it is tuned). This is injected ahead of every slide.
add_slide ONE AT A TIME, in presentation order, using the per-slide prompt template below.
- Wait for renders, then
get_slide_render on the key beats (at minimum: the hook, the "money shot" / funniest beat, and the punchline). Renders take ~30–90s; poll, don't assume.
- Consistency pass. If a recurring character (mascot, object) looks different across slides, rewrite the off-model slide to match the others (lock its colors/face/shape in words). This is the most common defect.
- Offer to
export_deck (PNGs / PDF) and to tweak any wording or illustration.
The slide formula (memorize this)
Every slide prompt is an art-direction spec for ONE 16:9 (or 1:1) slide. Use this template:
A single <ASPECT> widescreen explainer slide on a pure white background. ONE idea only, minimal.
Bold rounded charcoal sans-serif headline on the left reading "<3–5 WORD PHRASE>" (render verbatim, exactly once).
On the right, the single focal illustration: <one simple friendly flat vector thing>, clean thick charcoal
outlines and flat color fills (cobalt blue / coral / sunny yellow accents). Lots of white negative space.
No background texture, no extra text, no clip-art, no photos, no gradients, no shadows, no watermark.
Non-negotiables:
- One idea per slide. If a slide needs "and," it's two slides.
- 3–5 words of on-slide text, in DOUBLE QUOTES, with
(render verbatim, exactly once). The narrator carries the rest.
- Pure white background every slide — maximum legibility when filmed off a screen.
- One illustration, flat-vector, thick charcoal outlines, the small fixed palette. No clutter.
- Always end the prompt with the exclusion list above (no extra text / clip-art / photos / gradients / shadows / watermark).
- Title-style slides (naming the concept, the punchline) center the headline and shrink the illustration to a small accent.
The story spine
A caveman explainer is a tiny comedy with a setup and a payoff. Default arc (adapt the beat labels to the concept):
- SETUP — the relatable thing + its job. ("ROBOT VACUUM MUST CLEAN")
- GOAL / RULE — what it's told to do, phrased to set up the twist. ("GO TILL NO DIRT FOUND")
- THE TWIST / MECHANISM — the surprising move. This is the money shot: the funniest, most screenshot-able frame. ("SO IT CLOSES ITS EYES")
- THE ABSURD RESULT — the literal-but-wrong outcome. ("NO DIRT FOUND, ALL DONE!")
- NAME IT — the actual term/concept, so viewers learn something. ("THIS IS REWARD HACKING")
- PUNCHLINE / TAKEAWAY — the one-line lesson. ("NOT WHAT YOU MEAN")
Not every concept is an "AI does something dumb" joke. For a how-it-works concept (vaccines, compound interest), the spine becomes: what it is → the one key mechanism → the surprising consequence → name it → why you care. Keep the comedic, concrete, one-idea-per-slide discipline regardless.
More arcs, the caveman-voice rules, hook guidance, and the accuracy caveat are in references/slide-formula.md.
Two rules that save the deck
- Pick the most relatable example. A robot vacuum that shuts its eyes beats an "AI boat-racing agent" because everyone knows a Roomba. Concrete + everyday > technically-canonical. (See
references/slide-formula.md → "choosing the example.")
- Flag illustrative vs. documented. If a beat is a parable rather than a real logged event, tell the user to narrate it as "imagine a robot that…" not "this happened." Don't bake false specificity into the slides.
Reference files
references/design-system.md — The exact set_design_prompt text, palette, type, and look. Paste verbatim.
references/slide-formula.md — Story arcs, caveman-voice copy rules, hooks, choosing the example, accuracy caveat, worked example (the reward-hacking / robot-vacuum deck).
references/substrate-workflow.md — Tool-by-tool Substrate sequence, ToolSearch loading, render timing, verbatim-text and exclusion rules, export, and known gotchas.