The PDF is missing.

Sorry for that, I have fixed it now.

1. I think the first thing you should do is gather as many samples of Osmanya script as you can.
2. Then you need to try to identify what works and what doesn't.
3. Once you have determined the intended use of the typeface (body text, I suppose), you should start designing letters (and test them in actual words and paragraphs), applying what you learnt at 2.

Also, you would want to track down some of Hrant's writing about non-latin typefaces creation:
http://www.google.com/search?q=hrant%20non%20latin%20site:typophile.com

What are your opinions on the existing Osmanya fonts?

One can't design a successful typeface without a deep understanding of the letter forms. Most of us can't even read Osmanya, so seeing one example of an Osmanya font as shown on the Unicode charts is not nearly enough to give us a proper understanding of the forms. Each writing system that has been in use for generations has a certain vocabulary of normative forms, most often determined by writing implements and materials, be it reed stylus on clay or calligraphy brush on paper. And of course many scripts have an existing typographic tradition. All these have to be taken into account for a design to be successful for body text.

Before you think I'm being too negative, suppose you showed someone who has never seen the Latin alphabet the Unicode chart featuring the Latin range. With only this to go on, would that person be reasonably expected to create a successful Latin text font?

I agree, we should gather as many samples of Osmanya writing as possible—handwritten, printed, in a variety of styles—before we start designing anything.

I have not seen a single good Osmanya typeface, and very little Osmanya writing.

One approach to designing for such a script, i.e. one that was invented by a specific individual rather than historically evolved, is to research its invention and development. This gives you insight into the thought processes of the inventor, and tends to uncover exemplary forms. This is the approach that my colleague Ross Mills took to the design of his Inuktitut and Cherokee typefaces.

Another approach is to experiment with different writing tools and to keep practising making the shapes until you develop a natural feel for their proportions and how they can be varied.

The MOST important work is by Prof. Andrzejewski -- SOAS, University of London.

I have gathered some samples to see what works and what doesn't. Then after that, I have made the font look very similar to the samples I had. This works well but does not look very good when enlarged; I have "file down" the rough edges. Do you think this looks good?

Here is the font and the sample

A pdf would be better for judging.
What I can see is that some gliphs are noticeably darker than others.

Osmanya looks it might do better following a Georgian model, while simple letters like O are used across scripts, a number of the letters seem to have very strong parallels. Not saying copy it, but it might help provide another jump off point

If nobody has noticed, I put the PDF online a very long time ago. When talking about the Georgian model, did you mean Khutsuri and Mkhedruli? I have looked Georgian up and have come up with those two styles. Now I'm working on a new font (I still have not abandoned the other one). The new font is a bit taller and more "italic" than my other font.