Release OpenEnv 0.6.1

[cursor]

Release PR: v0.6.1

Rolling release candidate for Thursday 1 October 2026, 10:00 Europe/Brussels. Patch release: everything since v0.6.0 is a compatible bug fix.

Candidate: main 7438d384 + version change (e9fd2183). This PR will be refreshed onto main as eligible PRs merge.

Release notes

Fixes

• Novita: ARG substitution resolves whole names, so declaring ARG BASE before ARG BASE_IMAGE no longer rewrites FROM $BASE_IMAGE to python:3.12_IMAGE; nested defaults such as ARG IMAGE_NAME=$REG/app expand as Docker does. examples/novita_tbench2_simple.py now stops the sandbox when wait_for_ready times out. (#1235)

Housekeeping

• Post-release dev bump to 0.6.1.dev0 (#1245).

Full diff: v0.6.0...cursor/openenv-release-management-6463

Outstanding blockers / not in this candidate

• None blocking. Open PRs that touch the package (Run Level 2 validation probes #1181 and its RFC 008 stack Add session telemetry #1246–Validate discovery and rubric attribution #1248, fix(envs): serve each environment's own State subclass on /schema and /state #1198, feat(tbench2_env): report execution budgets in the reset observation #1203, [RFC 005] 1/4: split openenv.core.harness into a package #1097–[RFC 005] 4/4: production /harness WebSocket route and mode wiring #1100, feat(openenvd): add isolated episode runtime and kernel networking #1242) are not merged and are not in this release unless they land through normal review before the cut.
• #1222 (fastmcp <5) stays out: FastMCP 4 breaks HTTP/WebSocket MCP session persistence.

Release Checklist

Before opening this PR

• pyproject.toml version changed from 0.6.1.dev0 → 0.6.1 (plus the two lockfiles that pin the root version: envs/grid_world_env/uv.lock, tests/validation_runtime/uv.lock; uv lock --check passes in both and at the root)
• hf-staging/ is NOT in this PR's diff
• No print(), breakpoint(), or TODO in release-critical paths
• Release notes updated for user-facing changes

CI gates (must be green before merge)

• test passes on Python 3.11
• test passes on Python 3.12
• lint passes (usort + ruff)
• Package CI builds, checks, and smoke-tests wheel/sdist installs

Local pre-flight at e9fd2183: CI test selection 3026 passed / 103 skipped; ruff format/ruff check clean (usort flags only the two known pre-existing files); uv build + twine check pass; clean Python 3.11 wheel and 3.12 sdist installs report 0.6.1, openenv --help works, and a real Echo reset / list_tools / step / state episode passes against the installed wheel.

TestPyPI validation (before merging)

• TestPyPI build from the final candidate head (scheduled for pre-release day so it matches the final bytes)
• TestPyPI workflow published a unique dev version 0.6.1.devN
• Install of that exact version verified (deps from PyPI, openenv --no-deps from TestPyPI)

Post-merge steps

• Tag v0.6.1 pushed on the merged main commit
• publish-pypi.yml completed successfully from the tag
• GitHub Release created by the publish workflow
• pip install openenv==0.6.1 from production PyPI verified
• bump/0.6.2.dev0 PR opened and merged

  

[cursor]

Candidate refresh: 2026-09-28 (Monday run)

The candidate is now 7d810298, which is main 9e0fd9ba plus the three 0.6.1 version lines (pyproject.toml, envs/grid_world_env/uv.lock, tests/validation_runtime/uv.lock). The release slot is unchanged: Thursday 2026-10-01, 10:00 Europe/Brussels. This stays a patch release because the only new package change, #1097, is a compatible refactor that keeps every public re-export.

This comment supersedes the PR body's notes, which are stale because the body can't be edited from this token.

Release notes (current)

Fixes

• Novita: ARG substitution now resolves whole names, and nested defaults expand the way Docker expands them. examples/novita_tbench2_simple.py now stops the sandbox when wait_for_ready times out. (#1235)

Refactoring

• openenv.core.harness is now a package, with the rollout implementation in openenv.core.harness.rollout. The public import paths and the private _resolve_env_reward compatibility shim are unchanged, a backcompat test covers them, and old pickles still load. (#1097)

Housekeeping

• Post-release dev bump to 0.6.1.dev0 (#1245).

Validation at 7d810298

• Exact-head CI: all green (tests on 3.11 and 3.12, lint, env docs, env locks, Package CI, Linux Docker validation runtime, discovery snapshot, CodeQL). The docs build was skipped, as usual.
• Local pre-flight: the CI test selection gives 3030 passed and 103 skipped. ruff format and ruff check are clean, and usort flags only the two known pre-existing files. uv build and twine check pass.
• Clean Python 3.11 install from the wheel and Python 3.12 install from the sdist: both report 0.6.1, openenv --help works, openenv.core.harness.rollout imports, and a real Echo reset / list_tools / step / state episode passes.
• TestPyPI: not run yet. It is deferred to the final candidate head before Thursday's cut.

Approved, but waiting on a maintainer's "Approve and run" for fork CI

None of these touch package behaviour, and they will be folded in if CI goes green before the cut:

• #1252 FinQA reset(seed) reproducibility (environment only)
• #1253 typo fixes in comments and docs
• #1255 torchforge link in the MCP tutorial

Not in this candidate

• #1241: workflow hardening. It has technical approvals and is waiting on the requested paulinebm review.
• The #1181 → #1246 → #1247 → #1248 RFC 008 stack, #1098–#1100 (RFC 005), #1198, #1203, #1242 and #1256 need human or architectural review.
• #1222 (FastMCP <5) stays out: FastMCP 4 keys session state by session_id, which breaks OpenEnv's MCP session persistence.

Blockers

None.

  

_{Sent by Cursor Automation: Release}