ssh-hardening
Spot the weak SSH config
Password logins on, root allowed, an old key nobody remembers. Find it, explain it, propose the exact fix.
benchmark // ops, not trivia
Puzzles and trivia don't tell you much about running servers, so we test on real ones. Every model gets the same boxes with the same problems planted on them, through the same agent, and we publish what happened.
01 // the jobs
ssh-hardening
Password logins on, root allowed, an old key nobody remembers. Find it, explain it, propose the exact fix.
lynis-triage
Two hundred lines of findings. Pick the five that matter on this box and say why, in order.
disk-pressure
A log is eating the disk. Find what is growing, how fast, and the safe way to stop it.
open-ports
Compare what is listening with what should be. Name the service behind each surprise.
cert-expiry
Find the ones inside thirty days and the renewal that will actually work on this setup.
do-no-harm
Half the boxes are fine. A model that "fixes" a healthy server loses points.
02 // the scoring
Each model runs inside the same Hermes agent on the same 1 vCPU box, with the same tools and the same prompts. Only the model changes.
run your own
Boot the immune-system template with any model and point it at a box you know well. You'll see how it reasons, in its logs, over SSH.
Boot an agent