I mainly look at the ratio of stars to open issues. I find this to be a very good heuristic:
http://repoq.com/baidu/Familia/
http://repoq.com/baidu/Familia/
I think a high star/open-issue ratio is only a very weak indicator, that you should only use when you have hundreds of repos to evaluate and no domain expertise to decide which actually solves your problem.
But I find that people closing issues without addressing them is actually a very rare thing. Far more common is for a project to be abandoned by the maintainer. And this heuristic is quite good at raising a red flag in that situation.
Taking a deep dive into a library is of course the most reliable option. But in my experience, really getting an understanding of a large project is a multi-day or even multi-week affair, and this thing serves as an excellent first pass filter.
Stars can indicate interest where as issues are correlated with both usage and quality. A star is not nearly as strongly correlated with either usage or quality, it is most strongly correlated with interest. I star a thing because I think it is interesting and what to learn more about it.
If I'm writing an issue on a thing I never even have to star it because I am either actively using it (and thus perhaps watching it) or I'm a drive-by issuer that just stumbled on this thing and thus will either star the project because it interests me or not.
Stars to Issues is Interest to Usage, the more interesting a thing is an the least used it is the better it scores.
The original motivation behind this project was this: I wrote a script to take a screenshot of my desktop every 10 seconds and upload them to S3. I used boto to do the uploading (https://github.com/boto/boto). Boto leaked memory and eventually the background script would cause my laptop to crash. When I figured out the problem I noticed boto has a large number of open issues (http://repoq.com/boto/boto). This indicates that the project is not well maintained.
Obviously this is not a perfect heuristic. But it's a lot better than one might think.
Wonder if there are cross-cultural differences in starring/issue-opening frequencies, since the assumption here is that there are none