[misc] chore: bump version to 0.3.9

Merged
JO
4d ago
+17-10

What

Bump PACKAGE_VERSION from 0.3.8 to 0.3.9 and replace the Unreleased changelog with concise release notes dated 2026-10-27. The notes cover #458–#461 and #464 since the published v0.3.8 tag: source-digest compatibility and redaction fixes, agent start timestamps and OpenSandbox request timeouts, local eval dataset-row uploads, paged/summary eval downloads, and the final dependency security requirements. The development-only #463 update is included in the release base but omitted from user-facing notes.

Why

Prepare the next stable SDK release with a complete comparison against v0.3.8. The notes retain the required migration for old named local eval runs: finish with the original SDK, use a new run name, or use --fresh to archive and restart.

Dataset-row uploads and explicit --type summary downloads require matching platform support. Release coordination proceeds on the maintainer's confirmation to treat those platform changes as available; this PR does not establish live platform acceptance.

How to Test

  • .venv/bin/ruff check --no-cache ., .venv/bin/ruff format --check --no-cache ., .venv/bin/ty check --error-on-warning osmosis_ai/, and git diff --check passed.
  • uv run --offline --locked --all-extras --group dev --no-editable pyright --verifytypes osmosis_ai --ignoreexternal reported 100% completeness with only the documented Harbor harness_agent baseline and no unexpected errors.
  • PYTHONDONTWRITEBYTECODE=1 LITELLM_LOCAL_MODEL_COST_MAP=True .venv/bin/python -m pytest -q -p no:cacheprovider --no-cov --tb=short --maxfail=5 completed with 4,065 passed, 1 skipped, and one failure because the editable installation still reported 0.3.8. After refreshing it with uv pip install --python .venv/bin/python --no-deps --offline -e ., rerunning tests/unit/eval/local/test_runner_units.py passed all 132 tests, including the failed version-header check.
  • Release metadata, the changelog bullets, PR links, and the v0.3.8...v0.3.9 comparison were checked; applicable pre-commit hooks passed.
  • Initial CI for 6daac9d4 passed: full Python 3.12/3.13/3.14 test suites with the coverage gate, lint, type verification, release artifacts, both Harbor compatibility variants, and all nine isolated wheel-install scenarios. No local production build or live platform/sandbox acceptance run was performed.
  • After syncing the final dependency fixes and notes, tests/unit/docs/test_doc_links.py passed (2 tests), git diff --check and pre-commit passed. Latest release CI passed on 29659edb with the final dependencies; all 21 checks passed.

Checklist

  • PR title follows [module] type: description format (labels are derived automatically).
  • ruff check . and ruff format --check . pass.
  • ty check --error-on-warning osmosis_ai/ passes; uv run --no-editable pyright --verifytypes osmosis_ai --ignoreexternal has no unexpected errors (see CONTRIBUTING.md for the Harbor baseline).
  • pytest passes across the complete Python 3.12/3.13/3.14 CI matrix; no new tests are needed for this version/changelog edit.
  • Public API changes are documented.
  • No secrets or credentials included.