Let's decide on some development priorities

I find my experience here has been mostly solid, so I think you mostly have your priorities right. It's also fun to see experimentation with new features every now and then, so I'm not sure I would even like to see a shift towards focusing only on important stuff.

• Mastodon integration #101
• Better handling of video content #147 (I am also interested if PieFed can integrate better with PeerTube, for example an option to browse PeerTube channels)
• "Announce does not federate properly with Lemmy" #22
• Fixing the bug where shared images disappear
• Account deletion #126

I think there's a lot of potential in better integration with services outside of the Threadiverse, and this is something Lemmy has been notoriously bad at (and its developers seemingly uninterested in). Not that PieFed should become like Kbin and serve two separate functions, but there's a lot to be achieved in both ends by making the services more interoperable. There might be some sort of place for both boosts and hashtags, while leading @mentions might sometimes be justifiably hidden. More than important I guess I would find it fun.

As for what's urgent, ensuring core functionality and GDPR compliance seems like a fair priority.

• Account deletion #126
• Fix image bug

I don't know which issues are urgent, I guess the ones preventing people from using the platform, but I don't know about any of those. These are just the ones I care about sorted by the importance I give them:

• The votes don't show after reloading.
• The markdown area where the posts are made seems a bit small.
• Subscribe to anything. I believe that many users appreciate the ability to follow other users, similar to Twitter or Mastodon, and have a feed of new posts for the things they follow instead of an inbox. However, I don't like this approach because if you follow a few people with diverse interests, you'll end up with a mix of many topics in your feed. When you want to see something specific, it can be difficult to find. I think a better solution is to use content tagging and custom feeds. This way, users can have multiple feeds and follow different topics in each of the feeds. While this may make things more complex, the simplest thing for people might be to just follow and see it in one feed. A possible compromise could be to have a custom feed as the default for everything followed, and provide a way to select a custom feed other than the default one when subscribing to something.
• Saved posts I think it could be better with collections of posts to have different saved lists for different topics. There could be a dropdown menu or popup to search or select a collection on a combobox with existing ones to add the post or comment to an existing collection or create it if it doesn't already exist. There could be a default collection to save things with a single click, like people are used to. This also gives the option to make some collections public and follow other users' collections.
• Videos #138, #147

I'll come back to this later, but for now:

• Federating from local communities #4. You have this as #22, but I've never seen any problem with signatures. I'm subscribed to /c/playground from a server that just dumps out activity into a file, without even checking the signature, so I'd say it's a problem of not being sent at all (I see the activity from local users, but not from remote users). I've had success in test environments for this, but not piefed.social.
• Improving error messages #24. More get_requests should be surrounded by 'try ... except', so it doesn't crash / say 'oops, something has gone wrong' if, for example, you enter an unresolvable domain in that form.
• Subscribe to anything #20 is something people want.
• Mitigate impersonation attempts (related to #52). Piefed doesn't currently check that the domain in the id and the actor's domain match. See this post, where the 'original on lemmon.website' points to a bbc news article. Technically, communities can announce anything - we can check that it is the community actually sending it, but not whether what it announces actually happened. Matching domains helps, so users can verify for themselves.

I need to investigate setting up piefed in production, partly because I keep blowing through my ngrok quota, but mostly because many of the Issues on codeberg don't become apparent or relevant in testing.

I agree with you that getting the fundamentals right for what is currently here is way more important than trying to change it into something else (i.e. do a few things well, rather than lots of things badly)