Security and attack vectors

[iboB]

The system has become much more open with plugins. We should have security in mind, try to identify attack vectors and handle them.

Some concerns:

• A malicious plugin which has inference return model_data and uses "models" from ~/.ssh
• A non-malicious plugin that allows paths as op params may be used to read arbitrary files from the system
• Likely many more exist

So it makes sense to run plugins sandboxed, but their model assets should be specifically accessible.

We must prepare a whole document regarding security and plugin development.

We should be able to vet plugins in some way.