Release 0.10.2

[github-actions]

Release readiness review (v0.10.1 -> TARGET fa442d9)

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

Diff

v0.10.1...fa442d9

Release call:

🟢 GREEN LIGHT TO SHIP Additive fixes and behavior tightening with no confirmed regressions in diff review.

Scope summary:

• 56 files changed (+4023/-1952); key areas touched: tracing, MCP approval handling, Responses streaming request IDs, RunState serialization, tests, and docs (including translations).

Risk assessment (ordered by impact):

1. Callable MCP approval policy now requires approval when agent is omitted
   Risk: 🟡 MODERATE. Tool calls that previously proceeded without approval may now block awaiting approval when agent is not provided, affecting integrations relying on dynamic policies.
   Evidence: MCPServer._needs_approval returns True when policy is callable and agent is None; MCPUtil.to_function_tool clarifies behavior change.
   Files: src/agents/mcp/server.py, src/agents/mcp/util.py, tests/mcp/test_mcp_util.py.
   Action: Run uv run pytest -k mcp_util and verify callable-policy tools require approval when agent=None, while static policies remain unchanged; pass when tests green and behavior matches docs.

2. RunState schema bump with new serialized field
   Risk: 🟡 MODERATE. New RunState schema version 1.4 and request_id persistence may break downgrade flows (older SDKs cannot read newer snapshots by design).
   Evidence: CURRENT_SCHEMA_VERSION changed to 1.4 and serialized model responses now include request_id.
   Files: src/agents/run_state.py, tests/test_run_state.py.
   Action: Run uv run pytest -k run_state and confirm serialized outputs include request_id and supported schema versions still deserialize; pass when tests green.

3. Streaming responses now wrap and attach request IDs
   Risk: 🟢 LOW. Wrapper adds cleanup and request ID propagation; risk is limited to streaming lifecycle handling if custom clients lack with_streaming_response.
   Evidence: _ResponseStreamWithRequestId wrapper and fallback path for clients without with_streaming_response.
   Files: src/agents/models/openai_responses.py, src/agents/run_internal/run_loop.py, tests/test_openai_responses.py, tests/test_agent_runner_streamed.py.
   Action: Run uv run pytest -k openai_responses and uv run pytest -k agent_runner_streamed; pass when request IDs appear on terminal streamed responses and streams close cleanly.

Notes:

• Base tag and target derived from local tags and HEAD; no remote tag fetch performed.
• Assumed CI verification already passed for this target; no local tests run here.