The deck already exists. These are for deciding whether it says one thing per slide, whether the chart is honest, and whether it survives being read rather than presented.
Generating a deck is already well covered. Anthropic ships a skill for producing presentation files, it works, and writing a sixth version of it would be a waste of everyone's time. So nothing in this category generates a deck.
These are for the pass afterwards, which is the pass that rarely happens. A deck arrives with forty slides, each carrying three ideas, a chart with a truncated axis, and a colour scheme that disappears on a projector. The narrative problem is upstream of all of that and gets discussed last, if at all.
The specific knowledge here divides into three kinds. There is structure, where the useful finding is that a slide headline stating the conclusion outperforms a slide headline naming the topic, and where a deck built to be read is a genuinely different artefact from one built to be presented behind you. There is honesty in charts, which is a short list of specific manipulations that are easy to do by accident: a truncated axis, two axes chosen to make lines cross, a pie chart past the number of slices anyone can compare, and colour carrying meaning that nothing else carries. And there is the generation constraint layer, which is real specification detail: the units a presentation file actually measures in, the slide dimensions for each aspect ratio, why automatic text fitting does not survive being produced programmatically, and what breaks when a font is not embedded.
There is a fourth kind, which is the one people are most surprised to find in a deck review at all. A presentation is read by assistive technology in the order the objects sit in the file, not the order they sit on the slide, so a deck that reads perfectly to the eye can be announced in an order that makes no sense, and that order is stored per slide with no deck-level fix. The same file holds the alt text, the table structure and the reading order that decide whether the exported PDF is usable at all. That is the accessibility skill. Alongside it sits the notes and timing pass, which exists because notes that repeat the slide get read aloud, and because a deck whose meaning lives in the notes is a deck that will be misread the moment somebody forwards it.
What is not here. Design taste, template selection, and anything about how to speak. The first two belong in the frontend design category and the third is not a thing a file can teach you.
None of the six has been measured. That is stated on every card and at the top of every page, and the file that is most obviously testable, the chart honesty audit, names the material a fair test would need.
Written, reviewed and free to take. No run behind them, so no claim about what they do to an output. Each page says so at the top.
Measured means the skill was given a realistic task on real material, then the identical task was run again with the skill removed, five runs each way. Each output was graded alone, against a rubric written by someone who had never seen the skill, by a session that was not told the other arm existed. Whether it passed was decided by a rule written down before any run executed. Those pages carry the worst case, the median, the p value, and what the skill costs you as well as what it buys.
Not yet measured means exactly that. It is written, it has been read, it is free to take, and we have run no experiment on it, so we make no claim about what it does to an output. It is not a skill that failed. Skills that failed are not published at all, in either state, and their numbers are in the results table on the hub.
Measuring one skill properly costs roughly twenty model sessions. We are working down the queue and moving skills from the second group into the first. Back to all skills.