5 projects
shobench
A benchmark of agents from shared initial conditions: what they do, and when they choose to stop
shorep
Replication packages for the shojin program: the data and the registrations
shogym
Environments that meet an agent at the tool surface
shojin
Tooling for the shojin research program: investigations into agent self-improvement. Name reservation; the package arrives with the program's shared tooling.
hgym
Harness Gym — environments for agents whose policy boundary is the tool surface
