html customization

I've been dreaming of making an in-browser GUI-based html editor/creator, with the goal of making front-end web development as accessible as using InDesign or Illustrator (hopefully even more so!). I'm hoping to conduct some user research in the next couple weeks to answer a philosophical question that comes up at this point: is the interaction metaphor of the adobe creative suite (selecting tools from a toolbar, using them at the end of a mouse to adjust objects in the document window, pressing "w" to see the end-user view) more intuitive/natural than the HTML interaction model (typing into a text editor, along with CSS into a separate window, and switching to a whole different program (if not two or three!) to view the result)?
But in the meantime a more pressing question is: has someone already created this functionality?
I have a vague suspicion that HTML5 includes this functionality - in my opinion it should! (see my report on end-user programming: http://www.dqe1.com/files/End-UserProgramming6.pdf ) But from what I've seen it's only vaguely defined so far.

http://blog.whatwg.org/the-road-to-html-5-contenteditable
- "contenteditable", which refers to the kind of rich-text editing window I'm typing into now at blogger.com - is, today in 2010, fairly fully implemented and supported across browsers
- "designmode", which refers to what I've described (i imagine), does not exist.

http://dev.opera.com/articles/view/rich-html-editing-in-the-browser-part-2/
describes in detail how to implement that/this text editor. Playing with the most advanced one I could find around (at okcupid.com), there's 0 drag-and-drop or other mouse-enabled editing, and while the buttons up top created only lists, line-breaks, and photos, typing in custom html rendered just fine in the box. But it didn't save to the server!

http://ljouanneau.com/lab/html5/demodragdrop.html

I've got some ideas for how to make this work and I'll detail them in the next couple weeks, but for now here's a few of the issues I'll need to clarify before the interaction can be specified:
1. which attributes will be editable?
- definitely size, position, color, and all the font-y details covered in a rich text editor
- to what extent does that require supporting complex/arbitrary and confusing/abstract concepts like float, margins vs padding, inline vs block, etc?
- javascript behaviors would be great, but that's probably more of a version 2.0 thing.
2. To what extent is it safe to let users create their own html?
- user generated html can be used to hack/exploit whomever downloads it later, if not also the server on which it's being saved
- I know this is probably why the example I tested above didn't save to their server.
- obviously there are limits but (one of these days when I get into javascript programming mode) I know there are ways to ensure that some useful functionality is available without compromising the host or eventual viewers
3. what's the use-case?
- basically, all my designer friends who have asked me "could you help me make a website?"
- everybody who's paying $100s for dreamweaver to use it as an html editor

invisible data storage

I'm far more involved in customizing, maintaining, and learning more about my software than most people, but I got a glimpse of the casual user's experience today when I briefly thought firefox was doing more than I knew it could.
I had to reinstall my operating system a couple of times in the past few months, so I expect most of my browsing history to be missing - but when was searching for some PHP programming help, I saw a purple link that I was sure I hadn't visited in over a year. While I remembered fairly quickly that I had visited it more recently, I was thrilled by the thought that FF had backed up my history online.
Now ordinarily I'd be pissed that they were invading my privacy this way (in fact I generally download and store on my own hard drive most of the gmail/gdocs/calendar data I wouldn't want to become part of the public record) - but for this functionality I would be thrilled for Mozilla to track my browsing history and help me either find or avoid the sources I've used before, depending on my intent.
I think the primary difference is that I have a model for my personal data in calendars, documents, and email. I've had those totally under my control, and I know all the details of what it looks like and what's in it. But with browsing history, I have no concept of the database structure, the kinds of details that are involved (and which are left out), nor how/when/where I would use that data in any other context. I think this is what keeps me from being upset about it: it doesn't really seem like *my* data, and almost not like anything at all - thus if Mozilla can turn it into any sort of insight at all for me, I feel like I'm getting that for free!

references & bibliography

a friend of mine - an academic historian - once mentioned that there is software specifically for tracking sources, quotes, and footnotes while you're composing a paper, and printing out a properly linked and formatted bibliography. (this is probably hidden somewhere in one of msword's sub-sub-submenus)
For years I've been occasionally daydreaming about displaying a universal bibliography in the form of a network graph (I've never gotten very far because I can't decide whether to have the links show explicit references, quotations/stealing of text or phrases, stylistic influences, shared subject matter...) that connects all the books in the western cannon. But it would be far simpler to perform this operation for a relatively finite set of books like the bibliography for a single paper.
The program could draw from a database that already lists all the references in the bibliography of published books, or rely on the user to scan in and OCR a few pages from each of the books. The resulting web would give a quick visual sense of how broadly the writer has researched or what state the field is in - if all the references are highly interconnected, it's a well-defined field and/or the writer hasn't performed any great synthesis, while an unconnected graph would indicate original research and new insights (assuming the database the links are drawn from is complete and accurate).