Release 0.14.2

[github-actions]

Release readiness review (v0.14.1 -> TARGET c68f062)

This is a release readiness report done by $final-release-review skill.

Diff

v0.14.1...c68f062

Release call:

🟢 GREEN LIGHT TO SHIP No concrete release-blocking regression or compatibility break is evidenced in v0.14.1...c68f062d70be5ffaa7054f34b229d9d4a51b989a.

Scope summary:

• 109 files changed (+7376/-4367); key areas touched: sandbox runtime/path confinement (src/agents/sandbox/**, tests/sandbox/**), run-loop/tool metadata persistence (src/agents/run_internal/**, src/agents/items.py, src/agents/run_state.py, tests/test_tool_origin.py), new MongoDB memory backend (src/agents/extensions/memory/mongodb_session.py, tests/extensions/memory/test_mongodb_session.py), model/provider fixes (src/agents/models/openai_chatcompletions.py, src/agents/extensions/models/litellm_model.py), docs/translations, and version bump to 0.14.2.

Risk assessment (ordered by impact):

1. Sandbox path-grant confinement changes are broad and security-sensitive

   • Risk: 🟡 MODERATE. A defect here could allow unintended read/write access outside intended workspace roots.
   • Evidence: Core path validation and remote realpath checks were refactored across src/agents/sandbox/workspace_paths.py, src/agents/sandbox/session/base_sandbox_session.py, and src/agents/sandbox/sandboxes/unix_local.py, including new extra grant semantics and write-deny handling (WorkspaceArchiveWriteError paths, new helper exit-code handling).
   • Files: src/agents/sandbox/workspace_paths.py, src/agents/sandbox/session/base_sandbox_session.py, src/agents/sandbox/sandboxes/unix_local.py, tests/sandbox/test_workspace_paths.py, tests/sandbox/test_runtime.py, tests/sandbox/test_docker.py
   • Action: Run targeted confinement regression checks: uv run pytest -q tests/sandbox/test_workspace_paths.py tests/sandbox/test_runtime.py tests/sandbox/test_docker.py -k "extra_path_grant or validate_remote_path_access or normalize_path"; pass criteria: all selected tests pass with no new permission-escape failures.

2. Run-item/tool-origin persistence touched streaming, approvals, and state serialization

   • Risk: 🟢 LOW. Potential impact is metadata consistency rather than core execution flow, and dedicated coverage was added.
   • Evidence: New ToolOrigin/ToolOriginType wired through model response processing, approvals, generated items, and RunState serialization/deserialization; explicit tests added including legacy schema roundtrip behavior.
   • Files: src/agents/tool.py, src/agents/items.py, src/agents/run_internal/tool_execution.py, src/agents/run_internal/turn_resolution.py, src/agents/run_state.py, tests/test_tool_origin.py
   • Action: Validate metadata/compat behavior: uv run pytest -q tests/test_tool_origin.py tests/test_cancel_streaming.py tests/test_items_helpers.py; pass criteria: all tests pass, especially legacy schema and streamed exception assertions.

3. Temporal sandbox example now depends on upstream temporalio plugin instead of vendored patch

   • Risk: 🟢 LOW. This is example-scope, but example startup can fail if environment/plugin availability diverges.
   • Evidence: Vendored plugin files were deleted and patch step removed; dependency bumped to temporalio==1.26.0; example imports now assume temporalio.contrib.openai_agents is present.
   • Files: examples/sandbox/extensions/temporal/_vendored_plugin/* (deleted), examples/sandbox/extensions/temporal/justfile, examples/sandbox/extensions/temporal/temporal_sandbox_agent.py, pyproject.toml
   • Action: Run example import/startup smoke test in Temporal extra env: uv run --extra temporal --extra daytona --extra e2b --extra docker python examples/sandbox/extensions/temporal/temporal_sandbox_agent.py --help; pass criteria: command exits successfully without import errors.

Notes:

• BASE tag was selected from local tags only as requested: v0.14.1.
• TARGET used current HEAD: c68f062d70be5ffaa7054f34b229d9d4a51b989a.
• Working tree is clean (git status --short produced no output).
• Assumption: standard CI verification stack for this commit is authoritative; this report is diff-based risk analysis.