Hey,

I lost my entire Hermes setup three times.

Not because I didn't know better. But because every time I got something working — a complicated agent, a new skill, a perfectly tuned workflow — I'd break it again. Wrong tweak, corrupted memory, overwritten config. And then: silence. Dead Telegram bot. Start from zero.

After the third redo I finally asked myself: why am I not backing this up?

So I hacked together a daily cron that tars my ~/.hermes/ into iCloud. Simple, worked great. Except — 2 GB every day. Mostly caches and node_modules I didn't need. iCloud full, manual cleanup, pain.

That's when I realized: a backup script is easy. A good backup system is harder.

You need to decide where backups live, what to include, how often, how many to keep, how to test them — and how to restore when things go south.

So I wrote a protocol. Just a list of requirements. I give it to my agent, it reads it, builds the script for my exact system, sets up the cron, tests it, documents everything. Works on macOS, Linux, Windows.

I made it completely free if you want to check it out:
https://blog.holmebengt.com/post.html?id=backup-protocol