Release 0.14.1

[github-actions]

Release readiness review (v0.14.0 -> TARGET 1752df9)

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

Diff

v0.14.0...1752df9

Release call:

🟢 GREEN LIGHT TO SHIP Patch release contains targeted fixes with added regression coverage and no concrete release-blocking evidence in the diff.

Scope summary:

• 129 files changed (+5202/-865); key areas touched: runtime fixes in src/agents/run_internal/, src/agents/computer.py, tracing exporter sanitization, substantial test additions, docs/ref and translated docs expansion, and version bump to 0.14.1.

Risk assessment (ordered by impact):

1. Server-managed handoff behavior now enforces/auto-disables unsupported history rewrites

• Risk: 🟢 LOW. Existing unsupported combinations now fail fast or degrade to delta-only input, which may surface as new warnings/errors for previously misconfigured runs.
• Evidence: 83b38333 adds _resolve_server_managed_handoff_behavior to reject handoff_input_filter and disable nest_handoff_history when server-managed conversation is active; coverage added in tests/test_agent_runner.py and tests/test_run_impl_resume_paths.py.
• Files: src/agents/run_internal/turn_resolution.py, src/agents/run_internal/run_loop.py, src/agents/run_config.py, src/agents/handoffs/__init__.py, tests/test_agent_runner.py, tests/test_run_impl_resume_paths.py
• Action: Run uv run pytest tests/test_agent_runner.py -k "server_managed_handoff" tests/test_run_impl_resume_paths.py -k "server_managed_conversation_flag"; pass criteria: all selected tests pass and warnings/errors match expected assertions.

2. Streaming guardrail stop-before-tool-execution hardening

• Risk: 🟢 LOW. Concurrency ordering changes in streamed turns can regress tool execution timing if not validated under race conditions.
• Evidence: fa049a26 introduces _triggered_input_guardrail_result, queues tripwire result early, and injects before_side_effects guard in streamed turn resolution to raise before tool side effects; new race-focused tests added.
• Files: src/agents/run_internal/guardrails.py, src/agents/run_internal/run_loop.py, src/agents/run_internal/turn_resolution.py, src/agents/result.py, tests/test_guardrails.py
• Action: Run uv run pytest tests/test_guardrails.py -k "parallel_guardrail_trip_before_tool_execution or slow_cancel"; pass criteria: tripwire exceptions are raised and tool execution remains false in those cases.

3. Computer action modifier-key forwarding with legacy-driver compatibility path

• Risk: 🟢 LOW. Added keyword forwarding (keys) could break nonstandard drivers, but the implementation now filters unsupported kwargs with warnings and has explicit compatibility tests.
• Evidence: 90134802 adds keys forwarding for mouse actions plus signature-based filtering/dropping for unsupported kwargs; regression tests include legacy/non-introspectable/kwargs drivers.
• Files: src/agents/run_internal/tool_actions.py, src/agents/computer.py, tests/test_computer_action.py, examples/tools/computer_use.py
• Action: Run uv run pytest tests/test_computer_action.py -k "modifier_keys or legacy_driver or non_introspectable"; pass criteria: modifier keys preserved where supported and downgraded to warning-only behavior for legacy drivers.

4. OpenAI tracing payload sanitization for non-generation spans

• Risk: 🟢 LOW. Usage metadata is now dropped for non-generation spans on default OpenAI endpoint; custom endpoints retain prior behavior.
• Evidence: 2cef83b4 gates usage retention by span type and endpoint path; regression coverage differentiates OpenAI endpoint vs custom endpoint behavior.
• Files: src/agents/tracing/processors.py, tests/test_trace_processor.py
• Action: Run uv run pytest tests/test_trace_processor.py -k "drops_non_generation_usage_for_openai_endpoint or keeps_non_generation_usage_for_custom_endpoint"; pass criteria: OpenAI endpoint strips non-generation usage and custom endpoint preserves it.

Notes:

• Base tag derived from local tags per request: git tag -l 'v*' --sort=-v:refname | head -n1 => v0.14.0.
• Target commit set from local HEAD: 1752df9bda9733fe6dd8fa3c3a1a2f9cc29e6010.
• Local verification was not executed in this review pass; assessment is based on diff/commit/test-change inspection.