Some Feedback on the AI-assisted Quarto authoring and review workflow in Positron

[henningsway]

I have been using Positron extensively for Quarto work recently, including the Quarto skills, and the experience has become very good overall. This is not meant as a set of fixed feature requests or prescribed solutions, but as feedback from day-to-day use and a few areas that may be worth considering.

My workflow alternates between the file explorer and a three-pane layout with Posit Assistant on the left and the editor and viewer side by side. I often comment directly in Markdown files, describe an implementation approach in a plan, and then have the Assistant implement it. At the moment, I find Positron particularly useful for exploratory data analysis and for finishing reports. By contrast, I find AI-generated prose less helpful at the beginning of the writing process.

A few observations from this workflow:

• Figures and tables as important navigation markers. Figures and tables are major landmarks when reviewing a report, not just headings. Since they already have labels, captions, and numbering, it may be worth considering whether they could play a larger role in navigation and review.
• The Quarto outline does not always fit the active layout. The current placement is not especially useful in the layouts I tend to use. I found the RStudio approach—where the outline could be shown on the right next to the editor—more convenient. More broadly, while reviewing a rendered report, it is not always easy to move from something visible in the viewer to the relevant location in the source file. Tighter connections between the viewer, outline, and source could make report review more fluid.
• The outline could distinguish document structure from implementation details. Named Quarto chunks are useful during authoring, but they do not always help with document-level navigation. In some contexts, an outline that emphasizes the report structure rather than code-chunk names would be easier to use.
• Interactive GT output and the Quarto Viewer can compete for the same space. When running a code chunk interactively, especially one producing a GT table, its output can take over the same viewer context as the rendered report. I do not have a fixed solution in mind, but it would be helpful to preserve easy access to both contexts or make moving between them smoother. Possibilities such as a second viewer area, separate window, or simple back/forward navigation might be worth exploring.

Overall, the main opportunity I see is to make the relationship between the Assistant, the source document, and the rendered viewer feel more continuous—particularly during report review and revision.

[juliasilge]

Thank you for this detailed writeup! I really appreciate your thoughtful feedback and it's a useful map of where the Assistant, the source document, and the rendered viewer don't yet feel like one continuous surface.

Two questions would help us know a bit more about your workflows:

1. Do you mostly work in the source editor for these .qmd files? Or the visual editor?
2. Do you have inline output enabled (positron.quarto.inlineOutput.enabled)? Or are you executing in the console?

The reason for asking is that several known Assistant/Quarto gaps line up along those two axes:

• If you work in the visual editor, you may be hitting Assistant: Quarto docs not added as implicit context if they're in visual mode #9326, where Quarto documents open in visual mode aren't added as implicit context for the Assistant. That fix is planned for the next release.
• If you're running chunks interactively with inline output, the Assistant currently can't execute code in the document's live kernel or observe its outputs. That's tracked in Agents cannot execute or observe Quarto documents in inline execution mode (epic: positron.quarto context API) #14733, an epic also planned for the next release that proposes a positron.quarto context API. The current plan does defer visual editor support FWIW.
• Relatedly, Quarto: Add AI actions to the Quarto editor similar to the notebook editor #12838 covers bringing the notebook editor's AI actions to Quarto documents.

Some of your other observations, such as outline placement and docking, distinguishing document structure from code chunk names, navigating from the rendered viewer back to the source location, and interactive GT output competing with the Quarto Viewer, don't map cleanly onto issues we currently have logged. We can dig into those, with a bit more detail from you!

[henningsway]

Thank you!

I mostly use the source mode for editing (though I am generally also a fan of the visual editor (ideally looking for some kind of obsidian feel and also markdown better support (for all those ai-plans there are now) :D).

Also I am more a console person than a notebook/inlineOutput-Person (because I am used to the R/REPL-Experience and ). I can see that inline Output may solve the issue with the double use of the the Viewer for the document and intermediate results.

One other issue, that might be relevant: Click-to-source (reverse search) between HTML preview and .qmd source