| version | alpha | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| name | presentation-skill Deck System | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| description | Agent-readable design contract for clean, aligned PowerPoint decks generated by presentation-skill. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| colors |
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| typography |
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| rounded |
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| spacing |
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| components |
|
presentation-skill creates editable PowerPoint decks from structured outlines. The visual system should feel precise, editorial, and deliberate: strong title hierarchy, stable margins, clean content blocks, and no overlapping elements. Alignment and readability win over decoration.
Use one preset per deck. A deck should have a dominant background family, one support color, and one accent. Avoid generic blue unless the topic genuinely calls for it. Dark title and section slides should contrast with light content slides, or the deck should commit to a dark theme throughout.
Titles must visibly dominate body copy. Use 36-44pt for main slide titles, 20-24pt for section/card headings, 14-16pt for body text, and 10-12pt for captions and source lines. Letter spacing is always zero; do not use negative tracking.
Use a 16:9 canvas with disciplined side margins and repeatable gutters. Content starts only after wrapped titles and subtitles have reserved enough height. Body text is left aligned except for KPIs. Every content slide needs a visual anchor: image, chart, icon system, table, oversized number, or a strong two-column composition.
For reusable workspaces, capture the taste decision in design_brief.json
before writing outline.json: audience posture, cover archetype, grid policy,
container/card policy, and the rhythm-break plan. The outline should implement
that brief rather than cycling through every available variant.
Before rendering, choose a deck DNA that controls motif, pacing, density, and which variants are allowed. Good decks vary by argument, not by randomizing slides:
- Lab results dashboard: figure/table first, compact semantic fills, restrained navy header, captions and interpretation strips. Icons are usually wrong here unless they label workflow steps.
- Board risk memo: dark opener, status colors, risk matrix, timeline, owner table, and direct decision language.
- Product/investor reveal: one cinematic KPI or hero moment, asymmetric product/value cards, map or product visual, launch timeline, clear ask.
- Editorial report: masthead cover, warm paper/serif tone, artifact or concept imagery, fewer larger text blocks, deliberate whitespace.
- Civic science policy: map/data anchor, policy tradeoff matrix, accessible plain-language captions, source lines, and implementation table.
Use deck_style.header_mode, title_layout, title_motif,
section_motif, timeline_mode, matrix_mode, and stats_mode only to
reinforce the chosen DNA. Do not mix multiple DNA patterns in one short deck.
Cards may use a light shadow or fine border, but the deck should not become a stack of floating panels. Use depth to separate functional content groups, not as decoration.
Cards may use modest radii only when they do not have edge-attached accents. If a top/side accent rail, header strip, or flush overlay touches the card edge, use a rectangular card body so the accent aligns cleanly. Do not place thin accent lines above or below titles; that is a common low-quality generated-slide pattern.
Title slides must not all share one house template. Pick a cover archetype that matches the deck DNA: lab plate, command center, poster, masthead, light atlas, or split hero. Content slides use a dark header bar, clean lab header, or clearly reserved header stack. Generated-image slides must be standalone and labeled as generated, with prompt/model/purpose metadata visible so the slide can be deleted without affecting the deck narrative.
Avoid repeating four equal cards as the default rhythm. Use feature-left stats, policy bands, open quadrants, staggered timelines, or open editorial events when the content is not truly modular.
Timelines should not always be rail-and-card diagrams. Use simple report bands
for academic/ops updates, a chapter-spread when one milestone anchors the
story, and open events only when the spacing itself carries meaning. If the
slide is really a process, figure, or comparison, do not force timeline.
Rhythm breaks are optional, not mandatory. Use kpi-hero only when one number
or date genuinely carries the decision. If the deck is an academic update,
methods summary, or lab-results readout, a plain table, figure, standard slide,
or bottom takeaway box is usually better than a forced hero slide.
Use the lab-report preset or another restrained light preset. Credibility
comes from figure-first layouts and clean report slides, not decorative icons:
put plots, workflow screenshots, microscopy/gel/readout panels, tables, method
diagrams, or concise bullets in the main region. For simple academic decks,
prefer a white canvas with a measured heading, optional colored heading card,
footer rule, sources, and page number. Use a bottom takeaway box only when it
helps the presenter state the result without editing formatting manually. Use
9-11pt captions for assay/run/source metadata, navy headers, and semantic
red/green/blue/orange accents for genotype, control, pass/fail, or risk states.
Use variant: scientific-figure for 2-4 figure-panel slides with subfigure
labels and compact captions. Use variant: image-sidebar for one large figure
plus interpretation, and variant: lab-run-results for compact
result dashboards with multiple editable tables, semantic green/red/yellow
cell fills, small footnotes, and a short interpretation strip. Then use
table, flow, stats, and comparison-2col before generic card grids.
Icons are optional; use them for conceptual cards or timelines, not as
substitutes for evidence figures or result tables.
Flow diagrams are optional and should not be used as a generic rhythm break.
Use flow only when a method, architecture, assay path, or operating sequence
is itself the evidence. Keep visible rows to four boxes; if there are more
steps, split the process, summarize stages, or use a table/timeline. The
fallback Mermaid renderer caps rows at four boxes and balances long flows, but the
better design choice is often to avoid the diagram entirely.
Do choose a topic-specific palette, add visual structure to every content slide, keep source/provenance visible when useful, and run geometric plus visual QA before declaring a deck done.
Do not repeat the same layout three times in a row, center body text, mix presets inside one deck, use qualitative text as a giant KPI, force a hero slide when no single metric deserves it, or bury generated imagery inside evidence slides without disclosure.
For public-topic decks, use source-backed imagery when it strengthens the
argument. A small number of Wikimedia/CC images with visible source metadata
is better than many decorative visuals. When external assets are staged, keep
assets/attribution.csv and let the renderer add an editable Image Sources
slide unless a custom credits slide already exists.
The quality loop is source edit → build → QA → rendered review → source edit
again. Use scripts/visual_review.py after the normal QA gate to create a
contact sheet and flag orphan-word, title-wrap, safe-area, footer-clearance, and
variant-rhythm risks. Treat it as a punch list for source edits, not a reason to
patch the generated .pptx directly.