TsukiNi22/ai-utils / skills/ai/presentation
presentation
UnverifiedMake a presentation from a plan or an idea - a time (given, or estimated and confirmed) sets the number of slides; the format (PDF, one-file animated HTML, PPTX, several or all) and the style (several proposals for the subject, "White blueprint" by default, or copied from a PPTX / PDF / HTML the user gives) are asked through interactive-questions. One deck.json gives every format plus speaker notes. Use for "fais une présentation / des slides / un ppt / un pdf de présentation".
Presentation (plan or idea -> slides in the formats asked)
SKILL_DIR = the folder of this file. One source, deck.json, built into an animated web page (one HTML file), a PDF
(16:9, one page per slide), a PPTX and the speaker notes (<name>-notes.md), all in the same style.
Everything written for the user's tools (scripts, JSON keys, file names) is English; the slides are in the language of
the request (French here), code and comments in English.
1. What is given
- A plan: the user's own outline (a file, a message): keep its order and its words, an idea / line of the plan is not forced to be one slide (one idea may take two or three slides, three small lines may share one).
- An idea / a subject: write the plan first (5-8 chapters, one line each) and show it before the slides.
- A model to follow (a PPTX / PDF / HTML / image of theirs): the style comes from it (section 3).
- Ground it: read what the subject is about (the repository, the docs, the files named), and the earlier decks of
the user when they exist (
~/testing/ppt-tek,docs/,presentation/folders): their vocabulary, their level, what they already explained. Never invent a number, a command or a source: what cannot be checked is marked "to check" in the notes.
2. Ask (ONE interactive-questions call, French, 3 questions)
- Durée - when the user gave a time, use it. Otherwise estimate it from the plan (about 1 content slide per
minute, between 0.5 and 1: a talk of 20 minutes = 15 to 20 content slides, plus the title, the section dividers
and the end) and ask whether it fits: options "
<N> min, ~<S> slides(Recommended)", the same shorter, the same longer. The number of slides follows the time chosen; say it in the question. - Format - options:
PDF + HTML (Recommended)·PDF + HTML + PPTX (tout)·PPTX seulement· a custom mix through Other. The HTML is one self-contained file with build animations (the points appear one by one), the PDF is the same without motion (every step shown), the PPTX is editable (no animation) and carries the notes. Ask in the same question whether the file will be imported into Canva (or Google Slides / Keynote): then build with--canva(section 5b); the user can also say it in Other. - Style - exactly 4 options, in this order:
White blueprint (default)- recommended, always first, the label ends with "(default)".- One other catalogue style that fits the subject (the "when" column below), different in mood.
Sur mesure (sans template)- a look designed from scratch for this subject and this plan: its own palette, fonts, frame and background drawn from the subject's world, not taken from any catalogue theme (section 3, "Custom look"). Say in its description what it would look like (colours, mood) in a few words.Autre / custom- the user gives a file to copy ("copie le style de ", a PPTX / PDF / HTML / image) or describes the look in words; the free-text answer is read as such. Usepreviewto show a mini mock-up of each look (title line, colours, font) when it helps to choose.
- Several decks asked in one go: one question call for all, not one per deck.
- Unclear subject / audience / level: ask it in the same call (header "Public").
3. Styles
SKILL_DIR/templates/themes.json (and ~/.claude/presentation/themes/*.json, the user's own). Never hard-code a
colour in a slide: the theme is the only place of the look.
| Theme | Label | When |
|---|---|---|
blueprint | White blueprint (default) | technical, code, architecture, repositories, courses: an architect's plan (off-white, fine blue grid, blue ink, title block) |
paper | Clean paper | any subject, reports, school defenses: neutral, readable |
night-terminal | Night terminal | system, shell, security, low level: dark, monospace, green |
warm-keynote | Warm keynote | talks, pitches, ideas: cream, orange, rounded, big words |
slate-corporate | Slate corporate | business case, company, formal defense: slate and deep blue |
neon-dark | Neon dark | games, creative, AI: deep violet, magenta and cyan |
Custom look (Sur mesure) - the free, bold option: it is NOT limited to colours, a neon or a catalogue mood; the
subject decides everything. Do not start from a catalogue row, start from an idea (a concept taken from the subject:
a poster, a notebook, a terminal, a map, a comic page, a museum label, a blueprint of something else...) and push it:
flashy colours or quiet ones, big type, hard shadows, full-colour section pages, giant numbers, tilted cards, gradients,
patterns, drawn shapes (inline SVG in bgimg as url("data:image/svg+xml,...")), an unusual layout per kind.
Two files in the deck folder, then --theme-file theme.json:
theme.json: every key ofthemes.json(label,dark,bg,paper,ink,ink2,tx,dim,code,codetx,warn,ok,font,head_font,mono,head_transform,head_spacing,head_weight,radius,bgimg,frame,inner_frame,cartouche) +"css_file": "theme.css"(or"css": "..."). Extra colour keys are allowed and become variables everywhere:"acc1": "#0d7a55","acc2_line": "#c4561a"give--acc1/--acc2-linein the CSS and are resolved in the inline SVG of the PDF (two worlds in one deck: one accent per part, set per slide with#sN).theme.css: free CSS loaded after the layout (HTML and PDF): override anything (.k-title,.k-section,.col,.flow > li,.k-quote,#hud,@font-face, even transitions). Class names are intemplates/slides.css(the stage, its scaling and the outline shell:templates/deck-base.css).templates/custom-example/is a complete bold example ("Poster pop": cobalt, coral and yellow, hard shadows) to take the level of boldness from, not to copy. Rules that stay: text readable (contrast on every background, checked on the sheets), fonts that exist on the machine (fc-list), nothing leaving the slide (the selftest says it), no::marker { content }(it crashes the PDF engine: uselist-style-type). WeasyPrint (the PDF) also: drops a whole rule that uses:has(), makes an absolute::beforeof a grid container a grid cell (use flex there), and the layout rules.slide ul/.slide ol/.slide libeat a bare class (.road): write.slide ol.road. Every direct child of a slide isposition: relative(above the decor): a block placed withposition: absoluteagainst the slide needs its figure set back toposition: static(#s9 figure { position: static; }). The PPTX carries the colours, fonts and background colour of the theme only (not the CSS), the--canvabuild ignores the CSS: say it. Offer to keep the result with the user's own themes (~/.claude/presentation/themes/,theme.cssnext to it).
Copy the style of a file the user gives (or "le style de la présentation sur le repo C++"):
python3 SKILL_DIR/scripts/extract_style.py <model.pptx|pdf|png|html|css> --name my-style [--save]
It prints a theme (colours and fonts read from the file). --save keeps it in ~/.claude/presentation/themes/ for the
next decks. It is a first guess: build the example (templates/example.json) with --theme-file, LOOK at the sheets
(section 5) and adjust the JSON (ink, bg, font, bgimg, frame...) until it looks like the model. Offer to
set up a style once (--save) so that it is proposed in the next questions.
4. Write deck.json
SKILL_DIR/templates/example.json shows every kind. Keys: title, subtitle, lang, author, date, duration,
theme, slides (+ bg_angle, bg_fade for the per slide backgrounds, reveal: "click" by default = the
points / columns / steps / data-s blocks appear one per click on the web page; "auto" = everything appears when the
slide comes in, one after the other in a short entrance, one click per slide - for a pitch or when the user does not want
to click through; the slide's own entrance animations stay; a block of an svg slide marked data-click="1", "2"...
still waits for its click in both modes: the figures of a problem slide, a punchline). A slide: kind, title,
chapter (the bar of the web page), lead (one line under the title), notes (what is said aloud), bg (optional:
one colour, or two for a gradient: the slide's own background; the web page fades the colours from one slide to the
next instead of cutting, the PDF paints each slide's gradient, the PPTX its first colour, --canva a solid colour). The
slides are #s1, #s2... in the HTML / PDF: a theme.css can style one slide or a run of slides
(#s2, #s3 { --ink: var(--acc1); }).
| kind | Fields | Use |
|---|---|---|
title | title, subtitle, meta | first slide |
section | number, title, lead | a chapter starts |
bullets | points: text or {text, sub} | 3-5 short points, shown one by one |
columns | columns: {head, points} | compare 2-4 things |
table | head, rows | real columns of data |
code | code, lang, caption | a command or a few lines (never a whole file) |
steps | steps: {head, text} | a path in 3-4 stages, revealed one by one |
image | src (relative path), caption, alt | a screenshot, a photo, a QR code |
svg | svg, caption | a diagram drawn inline (currentColor = the text colour, <g style="color:var(--ink)"> for the accent, no fixed colour; the PDF resolves them from the theme); HTML / PDF only. The field is raw markup: an HTML block (a mock-up, a QR card, a custom layout) works too, and its data-s="1", "2"... are revealed one by one like the points |
quote | text, by | the sentence to remember |
end | title, subtitle, links, src | last slide: questions, links, QR code |
Text accepts `code` and **bold**. |
Colours in inline SVG (the bugs met, the builder fixes the first, check_deck.py warns on the others):
- A theme colour goes in
style, never in a presentation attribute:style="fill:var(--paper)", notfill="var(--paper)"(a browser paints it black).build_deck.pymovesfill/stroke/stop-color/colorattributes holding avar()intostyle, and merges twostyleattributes on one tag (the second would be dropped). - No
opacityon a<text>: WeasyPrint cuts or shifts it in the PDF. Use a dim fill (style="fill:var(--dim)"). - A point on a curve (a station on a line) is computed from the curve (
B(t)of the cubic, or a segment end), never placed by eye.
How a good deck is written (taken from the user's earlier decks and their remarks):
- Less text, more visual: a diagram, a table, steps, a command instead of a paragraph; what is said aloud goes in
notes, not on the slide. At most 5-6 points per slide, 60 words, one idea per slide. - Titles say the idea ("Un Makefile qui lance CMake"), not the topic ("Makefile").
- Examples from the user's real projects and real commands, tested when they can be run.
- A chapter order that answers the questions of someone new, in order; a recap or a quiz slide when the talk is a course.
- The PDF is read without the speaker: it must be understandable alone, the HTML is for presenting.
- A link worth keeping goes with a QR code image on the last slide (
image/endwithsrc). - No emoji, no decoration without meaning, nothing that moves unless it reveals the next point.
5. Build and check
python3 SKILL_DIR/scripts/build_deck.py deck.json -o <folder> --formats html,pdf,pptx,notes [--theme NAME | --theme-file F]
python3 SKILL_DIR/scripts/check_deck.py deck.json --dir <folder> [--sheets <tmpdir>]
check_deck.py: the pace (content slides per minute againstduration), too many points / words, missing notes or images, the number of PDF pages; then contact sheets (PNG of every page of the PDF and of the PPTX rendered by LibreOffice) and the selftest of the web page (headless Firefox: text out of the slide, hidden by the bar, clipped, the keys). Open the PNGs (Read) and LOOK: a slide that overflows, a table too wide, a title on four lines, black on black. Fix the JSON (split the slide, shorten) and build again. Put the PNGs in /tmp, not in the folder delivered.- The web page keys: arrows / space (next step), Home / End,
O(outline: chapter cards, the current slide marked, close with the cross or Esc;?mapin the address opens it),F(full screen),N(notes),H(hide / show the bottom bar), swipe. - Tools:
weasyprint(PDF),python-pptx(PPTX),firefox,pdftoppm,soffice(checks). A missing one is said with its install command; ask before installing (pip install --user python-pptx,sudo dnf install python3-python-pptx).
5b. Import into Canva (PDF or PPTX)
Canva rebuilds a PDF / PPTX as editable elements: it cuts a text where the font changes, drops a font it does not have, merges lines into one box, turns grouped shapes and gradients into one picture. So for a deck going to Canva:
python3 SKILL_DIR/scripts/build_deck.py deck.json -o <folder> --formats pdf,pptx --canva
python3 SKILL_DIR/scripts/check_deck.py deck.json --dir <folder> --canva
--canvakeeps ONE family per role (body, headings, code) chosen in the Canva library (templates/canva.json, installed here so that it is embedded in the PDF), no fallback fonts, no letter spacing, no gradient grid background, no background / padding on inline code, no auto-fit in the PPTX. The look of the theme (colours, frames, layout) stays.check_deck.py --canvareads the built files back: every string of every slide is found again as real text on its PDF page / in its PPTX shapes (a miss = text cut or turned into a picture), every font embedded and in the Canva list, no Type 3 font, no group or placeholder in the PPTX, no picture on a page that has none, no run without a font. An ERROR is fixed and rebuilt; a WARN is said to the user.- Prefer the PPTX when the text must stay editable in Canva (it keeps one box per text and the table as a table); the PDF
when the look matters more.
svgslides become pictures in Canva: useimageorcolumns/stepsfor what must be edited. - What no script can check is how Canva itself imports it: tell the user to open it once and compare with the contact
sheets; a font that Canva replaces anyway goes into
templates/canva.json/ out of it, then build again.
6. Deliver
- Folder: the one the user names, otherwise
presentation/<name>/at the root of the current repository (or of the current folder):deck.json,<name>.html,<name>.pdf,<name>.pptx,<name>-notes.md, the images used. - Tell in French: the files, the time and number of slides, the style, what was not verified (a command not run, a
figure to check), and what the PPTX does not carry (animations, inline drawings: give an
imagefor those), and for a Canva import that the import itself was not tried. - A change asked afterwards ("plus court", "autre style"): edit
deck.json, rebuild, no rewrite.
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: 196k | 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: 196k | 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: 196k | 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: 196k | Updated | Claude CodeCodex | Verified | Install |