This is a fun project you can do. Find yourself a good self-propagating support bot of a variety which hasnt been adapted for a virulent or damaging payload. The best things to look at are Malware that borders on Ad-Ware. Something a virus program wouldnt remove but something like an AdWare package would flag.

If you are into project lifecycle - take an extra step to model how long this little zombie-notifier will survive in the wild. If it is healthy in several versions of windows, or happy in a Cisco router, model how long those devices will be used without replacement or firmware updates.

Clearly stay well away from any headline-making bugs and stick to low priority things like command and control bots which dont have Spam or keylogger/data collection integral functions.

It used to be safe to just repackage CIA "Clumsyware"(TM) but now that they seem unable to keep track of their shit I would avoid it. (Look to the French? Macron has their people on top right now). But of course with the fucking French you have to basically learn French to use any of their shit.

So after you have deployed, start thinking about how the world wide web 'sees' you. I have some of my own so I can't be precise - but how many services off the top of your head are tracking "You" in a real way? Here is a freebie - something like Zillow would know if your property changed hands. Facebook would know how long it has been since you have been active. Voat would display parsable timestamps to any posts you made.

So now that your little reporting system has some target heuristic data to monitor, add a little e-mail notifier. Probably best not to over-complicate things here. You want the self-propagating elements to be able to deliver the message independently. For the sake of this completely hypothetical and experimental exercise - maybe the bots just post a key on 10-15 different social media sites. One or two of them can be on the darkweb assuming you have big enough numbers to exploit machines using such networks at a low level. (Eg not a self contained proxy like TOR+Browser)

Now about that key - it should be encrypted within the bots. So how to do that? You need to come up with some trigger patterns - things that would appear in your 'meta test trigger scenarios' if you disappeared/died. This is the single most time consuming test parameter. You need to find something that would create a very distinct, parsable pattern if you disappeared, and use that as a 'hash'/privkey for your bot to decode your 'unlock hash' before reporting it. My method is to have multiple instances of my 'master key' with various hashes (predictors of meta data expected upon your death or disappearance) and each of them are broadcast with each of my 'target' predicted hashes. To nail it you probably want to work out at least dozens if not hundreds of scenarios which means people will have to try dozens or hundreds of hashes.

Putting your dead mans letters is as simple as going remote and uploading some weird shit to the cloud and then posting around to the chans or reddit or sourceforge and making it clear it is an insurance file of sorts.

People like you and me, we hunt for these insurance files and learn how to open them when people disappear.