ericrisco/rsc-harness / skills/presentations
presentations
VerifiedUse when building, theming, or exporting a presentation deck — pitch, sales, keynote, board/QBR, leave-behind one-pager — from slide structure to a token-based theme to PDF or editable PPTX (Marp, Slidev, python-pptx). NOT the words (that is `marketing`), NOT the visual tokens (that is `design`), NOT the investor story arc (that is `pitch-deck`).
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 20.6 KB |
| evals/README.md | markdown | 2.8 KB |
| evals/cases.yaml | config | 6.4 KB |
| references/brand-grounding.md | markdown | 9.0 KB |
| references/markdown-decks.md | markdown | 11.8 KB |
| references/pptx-python.md | markdown | 10.6 KB |
| references/slide-design.md | markdown | 17.6 KB |
| references/storytelling-and-decks.md | markdown | 10.0 KB |
| scripts/verify.shexecutable | script | 9.7 KB |
references/slide-design.md
View on GitHub# Slide Design — Communication Design, Visual System, Data Viz & Motion
How a slide *communicates*, *looks*, and *moves*. The brand's visual system from `../design/SKILL.md`,
re-tuned for the worst viewing conditions a deck faces: a projector, a bright room, the back row, a glance.
Defer the token *system* (OKLCH palette, type pairing, spacing scale, motion personality) to `design`; this
file is the deck-specific application. Good/Bad contrasts throughout.
The first half is the **communication layer** — assertion-evidence structure, cognitive load, and what you
*show* vs *say*. The second half is the **visual layer** — grid, type, color, data viz, motion. Get the
communication right first; a beautiful slide that says nothing still fails.
## Design first: audience, message, evidence
Before a grid or a color, answer six planning questions. The deck serves the *audience's* understanding,
not the presenter's comfort — that is the one principle every other rule descends from.
1. **Who, specifically, is the audience?** Their existing knowledge sets the floor and ceiling for what you can assume.
2. **What is the ONE main message?** One sentence they should remember a week later. If you can't write it, the deck has no spine.
3. **What 3–5 supporting points** carry that message, and in what order do they build?
4. **What evidence** (a number, a chart, a demo, an example) proves each point?
5. **What action** should the audience take — the single call-to-action the deck climbs toward?
6. **What is essential vs. expandable** under time pressure? Mark each point so a 30-minute deck collapses to 10 without losing the spine.
This maps onto the deck arc and slide skeleton in `storytelling-and-decks.md`: the one message is the
thesis, the supporting points are the act beats, the CTA is the closing ask. Plan here, storyboard there.
## Assertion-evidence: the headline IS the slide
The single highest-leverage structural move in the whole skill. **Each slide is one assertion (a complete
sentence stating the point) plus the visual evidence that proves it** — never a topic label over a bullet list.
- **Assertion** = the slide title, written as a full claim: *"User engagement rose 43% after the redesign"*, not *"Engagement"* or *"Key findings"*.
- **Evidence** = the visual that backs it: the chart, the diagram, the screenshot, the one number. The body proves the headline; it does not repeat it.
- If a slide's content is a bullet list, ask whether it should be a slide at all, or whether each bullet is really its own assertion-evidence slide.
> **Good:**
> ```text
> Title: "User engagement rose 43% after the redesign"
> Body: [line chart, pre→post, the post point highlighted]
> ```
> **Bad:**
> ```text
> Title: "Key findings"
> Body: • Data shows an increase
> • Users engaged more
> • Revenue improved
> ```
## Cognitive load: one concept per slide
Working memory is tiny and the audience is also listening to you. Every slide must pass one test: *"What
is the ONE thing I want them to take from this?"* If there are two answers, it is two slides.
- **One concept per slide.** A slide that needs two breaths to explain is two slides.
- **Progressive disclosure** when a concept genuinely has sequential parts: show the initial state, then add element one with context, then element two building on it. Build order *is* the explanation — the audience looks where you are, not three bullets ahead. (This is the legitimate use of motion; see the Motion section.)
- **Text density is a ceiling, not a target** (see the limits table below). Less on the slide means more attention on you.
- **Code on technical slides:** always syntax-highlight; highlight the one critical line; build complex examples up; strip boilerplate that isn't the point.
## Spoken vs. shown
The slide and your mouth are two channels, not one. Redundancy wastes both. Reading the slide verbatim is
the fastest way to lose a room — they read faster than you talk, finish, and tune out.
| Put on the slide | Say out loud |
| --- | --- |
| The assertion (headline) | The elaboration and nuance |
| The visual evidence | The context and the how |
| The critical number | Its interpretation — why it matters |
| The next step / CTA | The story that earns it |
> **Good:** slide shows a chart and one claim; you narrate the backstory, the caveat, and the implication.
> **Bad:** slide holds your full script in bullets; you read it aloud; the audience read it 20 seconds ago.
## The core constraint: legible from the back of the room
A web page is read at 50cm. A deck is read at 3–10 metres, often off a washed-out projector, sometimes for 5
seconds before you click. Design for the *worst* case.
- **Body text ≥ 24pt; ≥ 28–32pt for a conference talk.** Title 40–60pt. Caption ≥ 18pt. Never shrink text to fit — split the slide.
- **~6 words per line, ~6 lines per slide** as a *ceiling*. Less is more. A keynote slide is often one line.
- **Contrast ≥ 4.5:1** for text (the WCAG AA floor), higher is better on a projector. Check both your dark and light themes.
- **One focal point per slide.** The eye should land in one place. If everything is bold, nothing is.
- **Negative space is not wasted space** — it's what makes the one thing readable. Crowded slides read as filler.
> **Good:** `OnboardingChurn-40%` in 54pt over a clean field, one supporting chart.
> **Bad:** a 14pt 9-bullet slide the presenter reads aloud while the back row squints.
## Layout grid for 16:9
- **12-column grid**, consistent gutters, generous outer margin (~7% / ~0.9"). Define a **safe area**; keep all important content inside it so nothing clips on a projector or in PPTX export.
- **One layout family**, reused: title, section divider, statement, content (headline + one visual), two-column (text|visual), full-bleed image, big-number, chart, quote, agenda/closing. Consistency reads as polish.
- **Align to the grid.** Ragged left edges and floating boxes read as amateur. Snap to columns.
- **Headline anchored top-left** (or centered for statements); visual occupies the focal zone; caption/source small and quiet at the bottom.
## Type for projection
- **Two type roles max:** a display face for headlines (personality) + a clean sans for body (legibility). Pull the exact pairing from the design tokens; don't introduce new fonts per deck.
- **A real type scale** (e.g. display 60 / h1 44 / h2 32 / body 26 / caption 18) — steps, not arbitrary sizes. One scale across the whole deck.
- **Weight for hierarchy**, not just size: bold the assertion, regular the support. Avoid all-caps for long strings (hard to read at distance); reserve it for short labels.
- **Numbers are the message** in data decks — set big numbers in display size; the label small beside them.
> **Good:** one display font for headlines, one sans for body, four sizes total.
> **Bad:** four fonts, random sizes, italic + underline + color all at once.
## Color: dark vs light decks, allocated for a room
- **Dark deck** (dark surface, light text): better for dark rooms, big LED/LCD screens, on-stage; reduces glare; feels premium. Default for keynotes/launches.
- **Light deck** (light surface, dark text): better for **printed handouts**, bright rooms, documents that double as leave-behinds, and projectors that wash out darks. Default for board/QBR leave-behinds.
- **Build both from the same tokens** (a `.dark` / `.handout` class in Marp/Slidev; two masters in PPTX). Keep contrast ≥ 4.5:1 in each.
- **60-30-10 allocation:** ~60% surface, ~30% ink/text, ~10% one accent for emphasis. The accent marks the *one* thing that matters per slide.
- **Never encode meaning by color alone** (color-blind audiences, washed-out projectors): pair color with label, position, or shape.
> **Good:** dark navy deck, off-white text, a single amber accent on the one number that matters.
> **Bad:** five competing brand colors per slide, a gradient behind every headline, low-contrast gray-on-gray.
## Data visualization: one chart, one point
The single biggest credibility lever in a data deck. **Every chart makes exactly one point, and that point is
the slide's headline.** If the chart needs a legend decoder ring, it's failing.
- **Name the takeaway in the headline** ("Revenue 3×'d in four quarters"), then let the chart prove it.
- **Pre-attentive emphasis:** gray out everything, highlight the **one** bar/line/point in the accent color. The eye goes there first.
- **Strip the clutter:** no gridlines (or one faint baseline), no chart border, no redundant legend for a single series, minimal axis ticks. Maximize data-ink.
- **Direct-label** lines/bars instead of forcing a legend lookup. Put the value at the end of the line.
- **Honest axes:** start bars at zero; don't truncate to exaggerate; one axis, not dual axes; consistent scales across comparable slides.
- **Pick the right chart:** trend over time → line; comparison across categories → bar; part-of-whole (≤ 4 parts) → a single bar or simple pie; correlation → scatter. **Never a 3-D chart, never a rainbow palette, never a pie with 8 slices.**
- **Big-number slide** for a single hero metric: the number in display size, one line of context, source in the caption. Often more powerful than any chart.
> **Good:** one line chart, four labeled points, the final point in accent, no gridlines, headline states the result.
> **Bad:** a 3-D stacked bar with 7 series, a rainbow legend, dual y-axes, and a title that just says "Metrics".
In Marp/Slidev, draw charts as SVG/chart-lib/images; in python-pptx use **native editable charts** (see
`pptx-python.md`) so the data lives in the file.
## Imagery
- **Full-bleed and intentional**, not decorative stock. One strong image > three small clip-art ones.
- **Always put a legibility scrim** (a dark/light gradient or solid overlay) behind text on a photo, or the text disappears against busy areas.
- **Resolution matters at 4K** — use images large enough not to pixelate on a big projector; prefer SVG for logos, icons, and diagrams (infinite scale, tiny file).
- **Consistent treatment** — one duotone/filter/crop style across the deck, from the brand imagery mood in the design tokens. Mixed photo styles read as a collage, not a brand.
- **Diagrams over screenshots** when explaining a system; a labeled, simplified diagram beats a dense UI capture nobody can read from row 10.
> **Good:** one full-bleed product shot, dark gradient scrim bottom-left, white headline over it.
> **Bad:** three pixelated stock photos of "teamwork", logos at random sizes, a screenshot too small to read.
## Motion & transitions — with restraint (a deck is not a video)
Motion clarifies sequence and state change; it does not fill time. Applies to HTML pipelines (Marp/Slidev
presenting) and to the build order you control in PPTX.
- **One transition family** for the whole deck (e.g. a quick `fade` or a subtle slide), ≤ 300ms. Mixing transition types per slide is noise.
- **Builds reveal one idea at a time.** Use `v-click` (Slidev) / fragment reveals to show bullets or chart series sequentially when sequence *is* the point — so the audience looks where you're talking, not ahead. Don't animate for the sake of it.
- **Animate to explain, not to decorate:** a value changing, a step being added, an old→new comparison. If the motion doesn't carry meaning, cut it.
- **Honor `prefers-reduced-motion`** in HTML decks — disable non-essential animation. (Irrelevant in the static PDF, which is why PDF is the safe leave-behind.)
- **Pacing:** statement slides reset attention after a heavy data slide; don't run ten dense slides in a row.
Borrowed-from-motion-tools ideas, applied to slides (NOT to turn this into a video):
- **Progressive reveal** of a diagram (show the current state, then the optimized state) — clarifies a transformation.
- **One highlighted element** entering/emphasizing — directs the eye, the slide equivalent of pre-attentive color.
- **Spring/ease on the single transition** for a premium feel — subtle, fast, consistent.
> **Good:** one fade transition deck-wide; a 3-step diagram builds one step per click; reduced-motion honored.
> **Bad:** spin/zoom/bounce transitions varying per slide, every bullet flying in from a different edge, 1.5s animations that make the audience wait.
## Content density limits (ceilings, not targets)
| Slide type | Limit |
| --- | --- |
| Title | 1 headline + 1 subtitle (+ optional tagline) |
| Statement | 1 line, big |
| Content | 1 assertion headline + ≤ 3 support points / 1 visual |
| Two-column | headline + text (≤ 4 points) \| one visual |
| Big number | 1 metric (display size) + 1 line context + source |
| Chart | 1 chart, 1 takeaway (named in headline), source caption |
| Quote | 1 quote + attribution |
| Image | 1 full-bleed image + scrim + ≤ 1 line of text |
## Diagnostic rubric (score 1–5, aim ≥ 4 each)
Use this to *evaluate* a deck — your own draft or one you're asked to review. Score each criterion 1
(failing) to 5 (exemplary); anything below 4 is a fix before shipping. The four axes match the QA gate
the `verify.sh` checklist points back to.
**1. Audience-centered design** — *does every choice serve the audience?*
| Criterion | 1–5 |
| --- | --- |
| Content matches the audience's knowledge level | |
| One clear main message + a value proposition for *them* | |
| Adaptable to time constraints (essential vs. expandable marked) | |
| Structure/order aids understanding rather than the presenter's notes | |
Red flags: presenter-focused not audience-focused; no read on existing knowledge.
**2. Visual clarity** — *does the eye land on the point?*
| Criterion | 1–5 |
| --- | --- |
| Assertion-evidence structure (claim + visual), not topic + bullets | |
| Visual elements balance text; one focal point per slide | |
| Visual hierarchy guides attention (weight, size, accent) | |
| Consistent layouts/treatment; deliberate whitespace | |
Red flags: bullet-point overuse; text-heavy slides; cluttered layouts.
**3. Cognitive load** — *can working memory keep up while listening?*
| Criterion | 1–5 |
| --- | --- |
| One key concept per slide | |
| Text density under the ceiling; slide ≠ script | |
| Animation judicious (disclosure/emphasis only) | |
| Code formatted; critical line highlighted (if applicable) | |
| Supporting detail accessible (appendix/notes), not crowding the slide | |
Red flags: multiple complex concepts per slide; text competing with speech; animation overuse.
**4. Accessibility** — *does it work for everyone, on any screen?*
| Criterion | 1–5 |
| --- | --- |
| Works across display sizes; reads from the back row | |
| Color contrast ≥ 4.5:1; meaning never color-only | |
| Inclusive imagery and language | |
| Font sizes appropriate (body ≥ 24pt; ≥ 28–32 for a talk) | |
Red flags: poor contrast; too-small fonts; non-inclusive content.
**Prioritize fixes:** (1) critical — blocks understanding, accessibility failure, message unclear;
(2) important — cognitive load, visual inconsistency, structure; (3) polish — animation, custom styling.
## Implementation checklist (build pass)
- **Structure:** main message clear in the first 2 minutes; supporting points in logical order; essential-vs-expandable marked; structure aids understanding.
- **Content:** assertion-evidence throughout; visual evidence backs each assertion; one concept per slide; code formatted and highlighted.
- **Visual:** consistent palette (3–5 colors); typography hierarchy (2–3 roles); deliberate whitespace; everything aligned to the grid.
- **Accessibility:** contrast verified; font sizes appropriate; alt text for key images; works across display sizes.
## Communication anti-patterns
These sit alongside the visual anti-patterns in `SKILL.md`; they fail at the *message* layer, before pixels.
| Anti-pattern | Why it fails | Fix |
| --- | --- | --- |
| **Data dump** — every slide raw data/charts, no interpretation | Audiences can't analyze in real time; without a stated conclusion they're doing your job; most is forgotten | One insight per slide; state the conclusion, show the proof; they grasp the point before the data |
| **Script reader** — slides hold the full spoken script as "bullets" | Audiences read faster than you talk, read ahead, then tune out | Slides show what you can't say; you say what you can't show |
| **Template trap** — content poured into a generic template | Design should serve comprehension, not just look professional; generic template → generic communication | Start from the communication need: what structure helps *this* audience understand *this* content? |
| **Animation circus** — transitions/builds/effects everywhere | Animation is attention; when everything animates, nothing stands out | Animate only for progressive disclosure or emphasis; default to none |
| **Bullet-point disease** — list after list as the default | Bullets are for documents; equal weight for unequal ideas; passive reading not active viewing | Assertion-evidence; if you truly need a list, question whether it earns a slide |
## Slide-design QA (the visual subset)
- [ ] Body ≥ 24pt (≥ 28–32 for a talk); reads at 3 metres; contrast ≥ 4.5:1.
- [ ] One focal point per slide; ≤ ~6 lines; generous negative space.
- [ ] Aligned to the grid; consistent safe-area; nothing clips on export.
- [ ] Two type roles, one scale, weight-for-hierarchy; no font soup.
- [ ] Colors from tokens, 60-30-10, one accent marks the one thing; meaning never color-only.
- [ ] Each chart makes one point named in the headline; decluttered; honest axes; no 3-D/rainbow.
- [ ] Imagery full-bleed + scrim, high-res, consistent treatment; SVG for logos/diagrams.
- [ ] One transition family ≤ 300ms; builds reveal meaning; reduced-motion honored (HTML).
- [ ] Dark/light variant chosen for the room (dark = stage, light = print/handout).
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| canvas-designanthropics/skillsCreate beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| docxanthropics/skillsUse this skill whenever the user wants to create, read, edit, or manipulate Word documents (.docx files) or Word templates (.dotx files). Triggers include: any mention of 'Word doc', 'word document', '.docx', '.dotx', or requests to produce professional documents with formatting like tables of contents, headings, page numbers, or letterheads. Also use when extracting or reorganizing content from .docx or .dotx files, inserting or replacing images in documents, performing find-and-replace in Word files, working with tracked changes or comments, or converting content into a polished Word document. If the user asks for a 'report', 'memo', 'letter', 'template', or similar deliverable as a Word or .docx file, use this skill. Do NOT use for PDFs, spreadsheets, Google Docs, or general coding tasks unrelated to document generation. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| pdfanthropics/skillsUse this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text/tables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages, adding watermarks, creating new PDFs, filling PDF forms, encrypting/decrypting PDFs, extracting images, and OCR on scanned PDFs to make them searchable. If the user mentions a .pdf file or asks to produce one, use this skill. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| pptxanthropics/skillsUse this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx or .potx file (even if the extracted content will be used elsewhere, like in an email or summary); editing, modifying, or updating existing presentations; combining or splitting slide files; working with templates (.potx), layouts, speaker notes, or comments. Trigger whenever the user mentions "deck," "slides," "presentation," or references a .pptx or .potx filename, regardless of what they plan to do with the content afterward. If a .pptx or .potx file needs to be opened, created, or touched, use this skill. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |