Release 0.8.4

[github-actions]

Release readiness review (v0.8.3 -> TARGET 2b4469d)

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

Diff

v0.8.3...2b4469d

Release call:

🟢 GREEN LIGHT TO SHIP No concrete release-blocking regressions found; follow-up validation suggested for new shell/MCP behaviors.

Scope summary:

• 43 files changed (+2459/-490); key areas touched: src/agents/ (tools, turn resolution, MCP, responses), tests/, examples/tools/, docs (including translations), dependency bump.

Risk assessment (ordered by impact):

1. MCP tool calls now pre-validate required parameters

   • Risk: 🟡 MODERATE. Existing callers that relied on server-side defaults for missing required fields will now fail client-side with UserError.
   • Evidence: New _validate_required_parameters invoked before call_tool in src/agents/mcp/server.py.
   • Files: src/agents/mcp/server.py
   • Action: Run uv run pytest -s tests/mcp/test_client_session_retries.py and confirm missing-parameter calls raise the expected UserError.

2. Shell tool hosted environment handling and shell_call_output normalization

   • Risk: 🟡 MODERATE. Shell calls are skipped for non-local environments; new environment validation may change execution paths and error timing.
   • Evidence: Environment normalization/validation added in src/agents/tool.py; hosted shell calls skipped in src/agents/run_internal/turn_resolution.py; shell tool payload includes environment in src/agents/models/openai_responses.py.
   • Files: src/agents/tool.py, src/agents/run_internal/turn_resolution.py, src/agents/models/openai_responses.py, src/agents/run_internal/tool_actions.py
   • Action: Run uv run pytest -s tests/test_shell_tool.py tests/test_local_shell_tool.py and confirm local shell still executes while hosted environments do not.

3. Prompt-managed tool_choice omission adjusted

   • Risk: 🟢 LOW. Tool choice handling changes only when prompt manages tools and tool_choice is a named dict; could affect edge cases relying on implicit tool choice.
   • Evidence: tool_choice now omitted only for dict choices in src/agents/models/openai_responses.py.
   • Files: src/agents/models/openai_responses.py
   • Action: Run uv run pytest -s tests/test_openai_responses.py -k tool_choice and verify expected tool_choice serialization.

Notes:

• BASE tag from local tags: v0.8.3. TARGET: 2b4469dcf50d700add43b84477f0d250b71c1446.
• No local verification executed here; assume CI ran the standard test suite.