Release 0.13.0

[github-actions]

Release readiness review (v0.12.5 -> TARGET 940c310)

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

Diff

v0.12.5...940c310

Release call:

🟢 GREEN LIGHT TO SHIP Version bump to 0.13.0 matches user-facing changes, and no concrete release-blocking regressions are evidenced in the diff.

Scope summary:

• 75 files changed (+3987/-590); key areas touched: src/agents/mcp, src/agents/models, src/agents/run_internal, memory/session handling, realtime defaults, and extensive targeted test additions.

Risk assessment (ordered by impact):

1. Realtime default model changed to gpt-realtime-1.5

• Risk: 🟡 MODERATE. Users relying on implicit realtime defaults will observe different runtime behavior/cost profile without changing code.
• Evidence: src/agents/realtime/openai_realtime.py switches default model from gpt-realtime to gpt-realtime-1.5; docs and tests updated in docs/realtime/* and tests/realtime/test_openai_realtime.py.
• Files: src/agents/realtime/openai_realtime.py, docs/realtime/guide.md, docs/realtime/quickstart.md, tests/realtime/test_openai_realtime.py
• Action: Run uv run pytest -q tests/realtime/test_openai_realtime.py -k "defaults_to_gpt_realtime_1_5 or session_config_defaults_audio_formats_when_not_call"; pass criteria: both tests pass and release notes explicitly call out the default-model change.

2. New MCP resource/session APIs are additive but expand surface area

• Risk: 🟢 LOW. API additions are backward-compatible, but new code paths can still hide edge-case transport/session behavior.
• Evidence: Added list_resources, list_resource_templates, read_resource, and session_id plumbing in src/agents/mcp/server.py; dedicated new tests added.
• Files: src/agents/mcp/server.py, tests/mcp/test_mcp_resources.py, tests/mcp/test_streamable_http_session_id.py
• Action: Run uv run pytest -q tests/mcp/test_mcp_resources.py tests/mcp/test_streamable_http_session_id.py; pass criteria: all tests pass with no flaky failures.

3. Concurrency/race fixes in tool execution and memory sessions

• Risk: 🟢 LOW. Changes target known race windows and are accompanied by focused regression tests, reducing but not eliminating concurrency risk.
• Evidence: Race-oriented fixes in src/agents/run_internal/tool_execution.py, src/agents/extensions/memory/sqlalchemy_session.py, src/agents/memory/openai_responses_compaction_session.py, plus new/expanded tests for eager task factory, cross-loop session writes, and compaction ID stripping.
• Files: src/agents/run_internal/tool_execution.py, src/agents/extensions/memory/sqlalchemy_session.py, src/agents/memory/openai_responses_compaction_session.py, tests/test_run_step_execution.py, tests/extensions/memory/test_sqlalchemy_session.py, tests/memory/test_openai_responses_compaction_session.py
• Action: Run uv run pytest -q tests/test_run_step_execution.py -k eager_task_factory tests/extensions/memory/test_sqlalchemy_session.py -k concurrent tests/memory/test_openai_responses_compaction_session.py -k orphaned; pass criteria: all targeted concurrency/compaction tests pass.

Notes:

• Working tree was clean during review.
• Base tag was selected from local tags only (git tag -l 'v*' --sort=-v:refname | head -n1 -> v0.12.5).
• Target was set to local HEAD (940c31076ab32c57b61acaea5119e4cb511bf884).
• Review is diff-based; no local full verification run was performed in this pass.