scaleapi/agentenv-framework released v0.9.1334. This change makes run_code pass the environment address to scripts running on the environment host regardless of provider, and separates each run into its own directory to isolate artifacts. According to the GitHub release record, this is the only change compared with the previous tag. The corresponding packages are agentenv-framework 0.9.1334 and agentenv-framework-protocol 0.1.349.
This repository presents a framework for building reinforcement learning (RL) environments. The repository description explains that building realistic RL environments requires researchers, engineers, and domain experts to collaborate on several aspects, such as artifacts, environment tooling, the variability of environments, and reproducibility. This release deals with how execution scripts receive the environment address in the process of building such an environment. It also includes the change of splitting directories per run, so artifacts are isolated on a per-run basis.
The public release information presents the change description and the package versions, but gives no performance figures or benchmark results. License information is also absent from this material. Therefore, it is not possible to judge how much execution performance or the overall environment-building process changes with this change. What can be confirmed is limited to passing the environment address and per-run directory isolation.