How to make Emacs more secure?

Hi all,

I started a job with quite strict security requirements, and I need to
reduce any risks, for example connected with possible injections of
malicious code.

For example, I will be running a coding agent harness in a virtual
machine to restrict its access to files on my host.

I'm considering doing the same with Emacs, but I'd really prefer to run
it on the host machine, for several reasons (GUI vs terminal is one of
them; also, I'd like to be able to use Magit, also to communicate with
the central repo, so it will need the credentials).

Now, what can I do to reduce risks connected with that?  An obvious
thing is to use as few external packages as possible, pin their versions
to known "good" ones, and of course restrict oneself to Elpa (so no
Melpa packages) and as few packages from Non-GNU Elpa as possible.

However, I'm not really a security expert, so maybe there are some risks
(or mitigations worth pursuing) I don't see?

I had a "chat" with an LLM, and here are its main points:

- There are risks connected with just opening untrusted files (file/dir
local variables, the enriched-mode issue we had some time ago, and many
more).

- Malicious agent skill could inject something into .git/, like
https://nvd.nist.gov/vuln/detail/CVE-2024-32002 (which I would not
notice because it is obviously not version-controlled).

- Malicious agent could even create a `.el` file with malicious content
and trick me into evaluating it.

Do you have any suggestions about how to deal with these risks?  Are
there any people in a similar situations who could share their (or their
companies') policies?  Any general opinions etc.?

TIA,

-- 
Marcin Borkowski
https://mbork.pl