ejtpd is kind of useless right now, being immature and difficult to debug. What I'm seriously considering is exposing an arbitrary code evaluation feature to authorized controller clients. This would allow much faster diagnostics, analysis, feature development, and a powerful administration fallback for when no other tools for the job exist. In fact, it could become the entirety of the protocol.
I am worried about security though, for obvious reasons. But the thing is, clients are authenticated administrators anyways. Every single specific concern I can come up with, I can easily shoot down with “if an attacker successfully masquerades as a controller, you're already totally fucked anyways.” What I'm wondering is, can anyone else think of any security concerns that can't be dismissed this way? If not, I think this is the most useful and development-easing feature we can possibly implement to improve ejtpd.