openenv integration

[willccbb]

Description

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Test improvement

Testing

• All existing tests pass when running uv run pytest locally.
• New tests have been added to cover the changes

Checklist

• My code follows the style guidelines of this project as outlined in AGENTS.md
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published

Additional Notes

---

Note

High Risk
Adds a large new sandbox-backed environment adapter and a new image build/push CLI that interacts with external services (Prime Sandboxes/registry), which can impact evaluation/training stability and infra usage if misconfigured.

Overview
Adds OpenEnv integration via new OpenEnvEnv (a MultiTurnEnv adapter) that boots an OpenEnv project server inside Prime Sandboxes from a required proj/.build.json manifest, generates seed-based train/eval datasets by resetting the remote env, and supports both gym (JSON action parsing) and MCP (tool-calling + OpenAI tool schema conversion) contracts.

Introduces a new vf-build CLI to build/push OpenEnv project images with prime images push, auto-detect contract from the server create_app(...) signature, and write/update proj/.build.json; prime env init gains --openenv to scaffold the enforced proj/ layout and templates. Documentation is updated to describe the new extra verifiers[openenv] and the vf-build flow, and two bundled example environments (openenv_echo, openenv_textarena) plus their embedded OpenEnv projects/manifests are added.

Misc follow-ups: bumps prime-sandboxes minimum version, centralizes TOML loading via load_toml, adjusts eval concurrency handling (grouped scoring uses group-level semaphore while the UI shows rollout-level concurrency), and skips the new OpenEnv examples in the generic environment smoke test.

^{Written by Cursor Bugbot for commit 07cdc76. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}