generate-openenv-env

Invoke the skill with a use case like "generate an env for the library textarena". It researches the target library's API, picks matching example environments as templates, scaffolds the envs/<name>_env/ directory, and implements models, client, server, and openenv.yaml.

The scaffold includes models.py, client.py, openenv.yaml, pyproject.toml, and a server/ directory with app.py, the environment implementation, and a Dockerfile. The openenv init CLI generates these with placeholders replaced, or they can be created manually from the template.

Use an MCP tool environment with MCPEnvironment and MCPToolClient when the use case centers on tool-calling interfaces rather than classic RL step/reset loops. The default typed EnvClient archetype fits most gym-style wrappers; specialized clients are only for cases standard clients cannot express.

Run openenv build and openenv validate --verbose from the environment directory, plus an import check with PYTHONPATH=src:envs. If no tests exist, start the uvicorn server and check the /health endpoint or run openenv validate --url against it.

No. The skill explicitly excludes training and evaluation tasks and only covers environment design and implementation. Its guardrails prevent routing into model training workflows and focus solely on building validated environment scaffolds.