Skip to content
Agent Orchestration
Skill

/deck-to-pptx

Build a PowerPoint .pptx file with python-pptx, on this deployment, without the deck engine. Use when the deliverable is a .pptx on disk: presentation, deck, slides, pitch deck, keynote, report deck, 幻灯片, 演示文稿, 做一份 PPT, 出一个 pptx, 汇报材料. Carries what is specific to this deployment

BOOST
From plugin
raven
5.3k23 skills11 agents
Install
$ npx -y skills add EverMind-AI/Raven --skill deck-to-pptx --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/deck-to-pptx

Context preview

The summary Claude sees to decide when to auto-load this skill.

Build a PowerPoint .pptx file with python-pptx, on this deployment, without the deck engine. Use when the deliverable is a .pptx on disk: presentation, deck, slides, pitch deck, keynote, report deck, 幻灯片, 演示文稿, 做一份 PPT, 出一个 pptx, 汇报材料. Carries what is specific to this deployment

SKILL.md

deck-to-pptx.SKILL.md
name: deck-to-pptx
description: "Build a PowerPoint .pptx file with python-pptx, on this deployment, without the deck engine. Use when the deliverable is a .pptx on disk: presentation, deck, slides, pitch deck, keynote, report deck, 幻灯片, 演示文稿, 做一份 PPT, 出一个 pptx, 汇报材料. Carries what is specific to this deployment -- which tools exist, which parameters they take, and what the gates refuse -- and nothing a competent author already knows. Pairs with design-editorial-and-presentations, which owns content order and editorial judgement. Not for a deck bound to one of the packaged templates: that route runs on Raven-PPT and uses ppt-script-authoring."

Deck to .pptx

The deliverable is a `.pptx` file on disk. Not a web page, an image, a PDF, or a Markdown outline. Name its absolute path in the reply.

What this deployment has

| Need | Call | Note | | --- | --- | --- | | Facts the material does not carry | `web_search`, then `web_fetch` the page | there is no other search | | A real logo, product shot, published chart | `image_search` | direct image URL, pixel size and source page per hit; see `references/assets.md` | | A picture that does not exist yet | `image_generate` | reference pictures go in `images`, up to six | | An icon | `raven_ppt.services.assets.icons` | 1304 outline icons, see `references/assets.md` | | A figure or table from a paper | PyMuPDF on `raven-python` | from the PDF you were given, else one you downloaded; crop the region or pull the embedded image; see `references/assets.md` | | A formula | `add_formula` from `raven_ppt.services.assets.formulas`, on `raven-python` | one line of TeX in, a picture in the deck's ink at true size out; see `references/assets.md` | | Symbols inside a sentence | `math_runs` from the same module | `_A`, `^2`, `θ*` become real sub- and superscript runs; see `references/assets.md` | | Render a page to look at it | `soffice --headless --convert-to pdf`, then `pdftoppm` | |

With no image key configured, `image_generate` says so. Say which pages would have had a picture and carry them on type, grid, rule and colour.

The pages are yours to draw

Every page is drawn here, with python-pptx: the grid, the title row, the type, the marks, the diagrams, the pictures and their treatment. Nothing about a page is settled by a default. `raven_ppt.services.assets` holds the icon set, `add_formula` and `math_runs`; take those and draw the rest.

On this route there is no ANTI-SLOP-CHECK.md, no contract and no Task State. Do not initialize or update one.

Read two skills first

`local/deck-to-pptx` (this file) and `local/design-editorial-and-presentations`, which owns the content order and the editorial judgement. Read both before the first page.

Settle four things first

Language, audience, length, ground. Take each from the first of these that settles it: the request; what memory recalled about this user; the defaults. Defaults: the language the request is written in; a general audience; about 20 pages; a light ground.

Ask with `ask_user` only for a decision none of the three settles and the deck cannot start without: at most one call, before any other work, recommending the default. Where there is no user to ask (the request came from another agent), take the defaults. Say in the reply which of the four were defaults.

Use what the user handed over as much as you can, before anything the rules below go and get: their documents and data are the material, their pictures are the pictures. A `.pptx` they attached is a template, and a deck on a template is the engine's work, not this skill's: the host routes it there before this skill is read.

Rules

1. Invent no number, no source, no person, no place. Search first. What the search does not find stays a gap that names what is missing and who supplies it. 2. A logo, a product screen, a real person, a real place, a published chart, a paper's own figure: **fetch it, never generate it.** A generated stand-in for something that exists is a fabrication the page presents as evidence. Generate only what has no original: illustration, backdrop, atmosphere. 3. Anything the deck names that has a face of its own -- a company, a product, a repository, a place -- has a picture somewhere. Search the whole deck's list in one pass before drawing anything: what the material names, and the ordinary furniture it never links -- the logos, the marks, the product shots. Keep each picture's page URL beside it for the source note. A page about five products with no mark of any of them is a page that did not look. 4. Decide the deck's visual direction once, before the first picture, and say what it is: ground, two or three colours, and whether pictures are photographs or drawings. **A brand's website is a starting point, not the verdict** -- a dark web hero does not make a dark deck, and a printed handout and a projected keynote want opposite grounds. Chaining every background off the first one gives a deck one look and no decision. 5. [layouts.md](references/layouts.md) holds reference shapes with their proportions, the type ramp, and nine compositions measured off the packaged templates. Read it to widen the list you choose from, not to pick from a menu: what a page has to say decides its shape. No one shape on more than 60% of the deck. 6. Repeating units -- a card, a row, a step -- take an icon from the packaged set. Search it by what the unit is about, not by a filename: 1304 of them ship beside this agent, and `references/assets.md` says how. An icon is a mark and not an illustration: about 0.7in, one weight and one colour across the deck, and no filled disc behind it. 7. The cover, the contents page, the closing page and every section opener get a picture with weight: a photograph found for it, or a generated background where no photograph says it. A real photograph of the subject beats a generated texture every time it can be had, so search

Read more
Ships withraven

One Surface, All Agents: Raven generates DAGs and orchestrates multiple specialized agents for complex tasks. Raven is the harness of harnesses, built for recursive self-improvement (RSI).

Get the whole plugin

Other skills on raven.

build-product-and-tool-interfaces
Skill

build-product-and-tool…

为浏览器内需要持续操作对象、推进多步状态并产生可检查结果的产品与专业工具设计、实现、重构或诊断完整界面。用于运营与审核工作台、任务流业务产品、创作与编辑工具、分析与管理工具、资产或文件处理…