Release 0.17.2

[github-actions]

Release readiness review (v0.17.1 -> TARGET 9e471be)

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

Diff

v0.17.1...9e471be

Release call:

🟢 GREEN LIGHT TO SHIP No blocking regressions, breaking API changes, data-loss risks, or release-critical packaging failures were identified in the v0.17.1...TARGET diff.

Scope summary:

• 40 files changed (+1727/-1000); key areas touched: docs/translations, src/agents/extensions/memory/, src/agents/models/, src/agents/run_internal/, src/agents/realtime/, src/agents/tracing/, and related tests.
• Release version bumped from 0.17.1 to 0.17.2 in pyproject.toml and uv.lock.
• Commit range contains targeted fixes for OpenAI Conversations reasoning persistence, tracing shutdown retry backoff, realtime unknown tool handling, approval rejection reasons, AsyncSQLiteSession settings, Chat Completions empty tool outputs, plus docs updates.

Risk assessment (ordered by impact):

1. No material risks identified

   • Risk: 🟢 LOW. The diff is consistent with a patch release: targeted bug fixes, tests added for each runtime behavior change, and no public API removals were found.
   • Evidence: git diff --name-status v0.17.1...9e471be194a12b5ef8abd4d48da7088cd4bb74ae shows modified runtime files and added docs only; no deleted public modules or renamed exports.
   • Files: pyproject.toml, uv.lock, src/agents/**, tests/**, docs/**
   • Action: Confirm CI passed make format, make lint, make typecheck, and make tests for TARGET.

2. Chat Completions tool-output conversion now substitutes placeholders for empty or non-text-only outputs

   • Risk: 🟢 LOW. This changes fallback behavior for an invalid Chat Completions representation, but preserves default compatibility by warning and sending [tool output omitted]; strict mode raises explicitly.
   • Evidence: src/agents/models/chatcmpl_converter.py adds _OMITTED_TOOL_OUTPUT_PLACEHOLDER and strict_feature_validation; tests/models/test_openai_chatcompletions_converter.py and tests/models/test_openai_chatcompletions.py cover placeholder, strict-error, mixed-output, and preserve-all-content behavior.
   • Files: src/agents/models/chatcmpl_converter.py, src/agents/models/openai_chatcompletions.py, tests/models/test_openai_chatcompletions_converter.py, tests/models/test_openai_chatcompletions.py
   • Action: Run uv run pytest tests/models/test_openai_chatcompletions_converter.py tests/models/test_openai_chatcompletions.py -q; pass criteria: all targeted tests pass.

3. OpenAI Conversations reasoning persistence handling changed

   • Risk: 🟢 LOW. The change preserves reasoning IDs/encrypted content for Conversations sessions and drops unpersistable reasoning items before sending to Conversations, which is aligned with server-managed persistence requirements.
   • Evidence: src/agents/run_internal/session_persistence.py preserves reasoning identity for OpenAIConversationsSession and filters reasoning without id or encrypted_content; tests/test_agent_runner.py adds coverage for preserving IDs, dropping unpersistable reasoning, and keeping encrypted content.
   • Files: src/agents/run_internal/session_persistence.py, tests/test_agent_runner.py, tests/memory/test_session_persistence_sanitize.py
   • Action: Run uv run pytest tests/test_agent_runner.py tests/memory/test_session_persistence_sanitize.py -q; pass criteria: all targeted persistence tests pass.

4. Tracing shutdown can interrupt retry backoff when a timeout is supplied

   • Risk: 🟢 LOW. The behavior affects shutdown latency during trace export retries and is covered by tests for timeout and no-timeout paths.
   • Evidence: src/agents/tracing/processors.py adds exporter shutdown signaling and interruptible retry wait; tests/test_trace_processor.py adds coverage for shutdown interrupting retry backoff and preserving full retries when timeout=None.
   • Files: src/agents/tracing/processors.py, tests/test_trace_processor.py
   • Action: Run uv run pytest tests/test_trace_processor.py -q; pass criteria: tracing processor tests pass without hangs or exit-code regressions.

5. Realtime unknown-tool handling no longer starts an automatic response

   • Risk: 🟢 LOW. This is a targeted behavior fix for unknown realtime tools and is covered by an updated test expectation.
   • Evidence: src/agents/realtime/session.py changes start_response=True to False for unknown tool output; tests/realtime/test_session.py asserts the new behavior.
   • Files: src/agents/realtime/session.py, tests/realtime/test_session.py
   • Action: Run uv run pytest tests/realtime/test_session.py -q; pass criteria: realtime session tests pass.

6. AsyncSQLiteSession now honors SessionSettings.limit

   • Risk: 🟢 LOW. The new optional constructor parameter is appended after existing parameters, preserving positional compatibility while aligning async SQLite behavior with other session backends.
   • Evidence: src/agents/extensions/memory/async_sqlite_session.py adds session_settings and resolves the default limit; tests/extensions/memory/test_async_sqlite_session.py covers defaults, constructor settings, explicit override, and limit=0.
   • Files: src/agents/extensions/memory/async_sqlite_session.py, tests/extensions/memory/test_async_sqlite_session.py
   • Action: Run uv run pytest tests/extensions/memory/test_async_sqlite_session.py -q; pass criteria: async SQLite session tests pass.

Notes:

• BASE_TAG was determined from local tags only using git tag -l 'v*' --sort=-v:refname | head -n1: v0.17.1.
• TARGET was determined with git rev-parse HEAD: 9e471be194a12b5ef8abd4d48da7088cd4bb74ae.
• Working tree status was clean.
• Local verification was not run as part of this review; this report assumes CI validates the mandatory release checks for TARGET.