Food-Ppt — Editable Presentations from Food & Nutrition Reports
SkillDocs & knowledgeTurn a food & nutrition review report, evidence brief, or manuscript into a fully EDITABLE PowerPoint (.pptx) that opens in PowerPoint / Keynote / Google Slides with every title, bullet, table, and figure as a native editable object, nothing flattened to an image. Converts the outputs of food-research, food-deep-research, food-paper, and food-pipeline (or any Word/PDF/Markdown/txt source) into a well-designed deck: it plans the outline, writes concise grounded slide text with presenter notes on every slide, places figures as movable picture objects, builds the .pptx via python-pptx, and QA-checks editability. Design informed by taste-skill. Use to make slides, a presentation, a PPT, or a deck from a report or paper. Triggers: make a PPT, create slides, presentation from my report, turn this review into slides, build a deck, PPTX from my paper, journal-club slides, conference presentation.
Use Food-Ppt — Editable Presentations from Food & Nutrition Reports in Claude, ChatGPT or Ahel Desktop
Free. Sign in, add Food-Ppt — Editable Presentations from Food & Nutrition Reports and connect your AI. About a minute.
Also: Claude Code · Cursor · Codex
Then ask your AI: use the Food-Ppt skill
Details
Instructions available. Your AI can read the instructions. Execution depends on the setup they require.
Account requirements not reviewed. Check the skill instructions before use; Ahel provides instructions and does not run this skill.
No other account needed.
Add Ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.
What this skill tells your AI
The instructions your AI receives, as published by pangenomeai/academic-skills-food-nutrition in food-ppt/SKILL.md and read by Ahel’s review.
Turn a review report, evidence brief, or manuscript into a fully editable
.pptx — every title, bullet, table, and figure is a native PowerPoint object the
author can edit, never a flattened image. Original work; the schema→builder→QA
architecture and design sensibility are informed by the open-source (MIT) ppt-master,
GordenPPTSkill, and codex-ppt-skill skills and the taste-skill and
ui-ux-pro-max-skill design guides (see the repo README Acknowledgements). No
third-party templates are bundled.
What it converts
The outputs of the suite's other skills, or any report/paper the user supplies:
food-reviewpanel report / Review & Response Report → a review-findings deck.food-research/food-deep-researchevidence brief or literature review → a synthesis deck.food-papermanuscript (or sections) → a paper-presentation / journal-club deck.food-pipeline— offered at FINALIZE to package the finished work as slides.
Required content for a proposal / paper deck
- Experimental-design flow diagram — a
flowslide near the front, built fromfood-figure's blueprint (experimental-flow.md) and rendered as native, movable PowerPoint shapes (one box per step, one arrow per transition — never a flattened image), so the audience grasps the design at a glance. Mandatory for any deck from a research proposal or research paper. - Results slides illustrate the result — each is a
resultslide: a figure (fromfood-figure, theme-matched) where feasible, or a scientific table when the numbers are the point, plus a Key findings panel of takeaways. - Big numbers and parallel facts get their own design — a single striking figure
goes on a
metricbig-number hero; a set of parallel facts (objectives, contributions, study parameters, KPIs) goes on acardsaccent-bar grid, not a bullet list (references/slide-design.md). - Executive summary — a closing summary slide is required in every deck.
- Theme-matched figures — figures use the deck theme's palette so they match the slides.
Editable-first (non-negotiable) — references/editability-and-qa.md
Build with python-pptx so the deck is genuinely editable: real text frames
for titles/bullets, native tables (editable cells) for data, and figures placed as
movable/resizable picture objects. Never render a slide to a background image,
never bake text into a picture, never deliver a PDF in place of a .pptx. If a figure
exists only as a raster, embed it as a picture object (still movable/replaceable) and
say so; prefer a vector/redrawable source where available.
Subagents (dispatch via the Agent tool)
ppt_coordinator— takes the source report/manuscript + audience/length, runs the flow, returns the.pptx+ a slide-by-slide summary.outline_planner— reads the source and builds the slide outline (references/source-to-slides.md): one idea per slide, a clear narrative arc, target slide count for the audience.slide_writer— writes concise, grounded slide text (bullets ≤ ~1 line each), speaker notes carrying the detail; every claim/number traces to the source — no new facts, no invented data.figure_placer— selects figures/tables from the source (or callsfood-figure) and maps them to slides as editable picture objects / native tables.deck_builder— assembles the deck-spec JSON (references/deck-spec.md) and runspython3 scripts/build_pptx.py <spec>.json --out <deck>.pptx.deck_qa— verifies editability, checks text overflow / slide density, and confirms every slide's content is faithful to the source (references/editability-and-qa.md).
Templates & design (references/templates.md, references/slide-design.md)
Pick one of 17 editable themes — slate (default) · sage · burgundy · teal · graphite
· ocean · terracotta · forest · consulting · dashboard · azure · geometric · indigo ·
scholar · claret · apricot · midnight (a dark theme) — with the deck-spec "theme"
field (catalog: templates/INDEX.md; preview: templates/theme-previews.svg). Most
recreate the look of the open-source GordenPPTSkill gallery (in English) as original
theme definitions; midnight recreates a Grok-generated dark deck. ppt_coordinator
auto-picks a theme from the deck's text via scripts/suggest_theme.py and offers the
top few as options for the user to choose. A theme is a palette +
type + layout style (cover brand bar + side panel, full-bleed section dividers, accent
edge stripe, underlined titles, big-number heroes, accent-bar cards, footer + slide
numbers, styled tables) applied as editable shapes — no bundled template files,
nothing flattened. ppt_coordinator chooses the theme from topic/audience or asks.
Restrained academic look informed by taste-skill and ui-ux-pro-max-skill: a real
type scale (strong size + weight contrast), one accent + ink/greys, whitespace, one idea
per slide, and layout variety — never the templated "AI deck" look.
Grounding & privacy (inherited)
Slides are only as true as the source — never add a finding, number, or citation
that isn't in it (food-paper/references/faithfulness-and-citation.md). Apply the
academic-style + human-writing.md pairing to slide prose. Run
scripts/privacy_scan.py on the deliverable; keep the mandatory AI-use disclosure
on a closing slide when the deck presents AI-assisted work.
Requirements
python-pptx (pip install python-pptx). Figures come from the source or food-figure.
Signals
- GitHub stars
- 33
- Forks
- 3
- Last commit
- Aug 2026
Ahel review
K1binfo
installs-packagesK1binfo
installs-packages (in agents/deck_builder.md)
Automated review, not a security audit. Ruleset v1+k2.
Advanced
- Item type
- skill
- Key
food-ppt- Source
- github.com/pangenomeai/academic-skills-food-nutrition
github.com/pangenomeai/academic-skills-food-nutrition
More in Docs & knowledge
Skill · mattpocock
More in Docs & knowledgecanvas-design
Skill · anthropics
More in Docs & knowledgedoc-coauthoring
Skill · anthropics
More in Docs & knowledgepopups
Skill · coreyhaines31
More in Docs & knowledgewriting-for-agents
Skill · mattpocock
More in Docs & knowledgespec-driven-development
Skill · addyosmani
More in Docs & knowledge