I’ve been experimenting with robot assistants in a similar fashion to our wonderful host. One thing that I’m wary of, though, is prompt injection attacks. So naturally I vibe-coded an Obsidian plugin that might help protect against these possible threats.
It’s nice because it’s simple. It just scans markdown files for sneaky text, strips that text, and tells you what it did. It can also quarantine non-markdown files so you can double-check those as necessary.
Obviously this is not an issue if you’re only ever writing your own text in your vault — but if you forget and do something dumb in four years, this’ll hopefully add a layer of security to limit the effects of your forgetfulness.
I haven’t submitted it to the Obsidian plugin gallery, so you’ll have to install it manually.
Obvious disclaimers apply: do your own due diligence; your mileage may vary; I may not be held liable, culpable, or closely; no motorcycles after 3pm.