Release 0.14.8

[github-actions]

Release readiness review (v0.14.7 -> TARGET 24ba93a)

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

Diff

v0.14.7...24ba93a

Release call:

🟢 GREEN LIGHT TO SHIP No blocking regression, breaking API change, packaging issue, data-loss risk, or security-impacting issue was identified in the v0.14.7...TARGET diff.

Scope summary:

• 8 files changed (+198/-10); key areas touched: MCP package re-export import behavior, sandbox instruction formatting, one realtime comment typo, package version bump to 0.14.8, and focused tests for MCP and sandbox behavior.
• Commit range reviewed:
  • 05004336 fix: agents.mcp masks the real import error when importing MCPServerStreamableHttp #3046 preserve MCP re-export import errors (fix: #3046 preserve MCP re-export import errors #3048)
  • 0661c9e9 fix: SandboxAgent custom instructions render inside SDK shell-command guidance #3043 delimit sandbox prompt instruction sections (fix: #3043 delimit sandbox prompt instruction sections #3047)
  • 572c7bf1 docs: fix typo in comment for WS event handler (docs: fix typo in comment for WS event handler #3050)
  • 24ba93a3 Bump version to 0.14.8

Risk assessment (ordered by impact):

No material risks identified.

1. MCP re-exports now use lazy import error preservation

   • Risk: 🟢 LOW. This improves optional import behavior while preserving actionable failures when users access unavailable MCP server exports.
   • Evidence: src/agents/mcp/__init__.py replaces eager server/manager imports wrapped in broad ImportError suppression with __getattr__ lazy exports, and tests/mcp/test_mcp_imports.py covers both package import success and chained import error preservation.
   • Files: src/agents/mcp/__init__.py, tests/mcp/test_mcp_imports.py
   • Action: Run uv run pytest tests/mcp/test_mcp_imports.py; pass criteria: both MCP import tests pass.

2. Sandbox instructions now add explicit section delimiters

   • Risk: 🟢 LOW. Prompt text shape changes by adding headings, but ordering is preserved and covered by updated tests.
   • Evidence: src/agents/sandbox/runtime_agent_preparation.py wraps agent, capability, and remote mount policy fragments in titled sections; updated tests assert expected headings and ordering before filesystem instructions.
   • Files: src/agents/sandbox/runtime_agent_preparation.py, tests/sandbox/test_runtime.py, tests/test_sandbox_runtime_agent_preparation.py
   • Action: Run uv run pytest tests/test_sandbox_runtime_agent_preparation.py tests/sandbox/test_runtime.py -k sandbox; pass criteria: sandbox instruction composition tests pass.

3. Release metadata bump is consistent

   • Risk: 🟢 LOW. Version metadata changes are limited to the expected 0.14.8 bump.
   • Evidence: pyproject.toml and uv.lock both update openai-agents from 0.14.7 to 0.14.8; no dependency versions or Python requirements changed.
   • Files: pyproject.toml, uv.lock
   • Action: Run uv run python -c "import importlib.metadata as m; print(m.version('openai-agents'))" in the synced environment; pass criteria: reported version is 0.14.8.

Notes:

• Latest release tag was determined from local tags only with git tag -l 'v*' --sort=-v:refname | head -n1: v0.14.7.
• TARGET was determined with git rev-parse HEAD: 24ba93a39c99d9c59c5d2ab88d868038db69c67d.
• Working tree status was clean.
• Local verification was not run as part of this review; this report assumes CI verification for TARGET is handled separately.