Python: Fix Anthropic option conflicts and manager parse retries

[eavanvalkenburg]

Summary

• fix Anthropic options handling by removing unsupported top-level kwargs (allow_multiple_tool_calls, stream) from prepared request payloads
• keep stream mode explicit at Anthropic SDK call sites to prevent duplicate stream kwarg conflicts
• add regression tests for Python: [Bug]: AnthropicClient TypeError: AsyncMessages.create() got an unexpected keyword argument 'allow_multiple_tool_calls' #3371 and Python: Streaming conflict when using Anthropic Skills with HostedCodeInterpreterTool #3827 scenarios

Validation

• uv run pytest packages/anthropic/tests/test_anthropic_client.py packages/orchestrations/tests/test_group_chat.py -q

Fixes #3371
Fixes #3827

[markwallace-microsoft]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/anthropic/agent_framework_anthropic
_chat_client.py	368	148	59%	422, 451, 483, 485, 504, 526–529, 538, 540, 577–581, 583, 585–586, 588, 593–594, 596, 629–630, 639, 641–642, 647, 664–665, 709, 728, 743, 745, 749–750, 793–795, 797, 810–811, 818–820, 824–826, 830–833, 844, 846, 868, 878, 900–906, 913–914, 922–923, 931–934, 941–942, 948–949, 955–956, 962, 970–972, 976, 983–984, 990–991, 997–998, 1004, 1012–1015, 1022–1023, 1042, 1049–1050, 1069, 1091, 1093, 1102–1103, 1109, 1131–1132, 1138–1139, 1148–1158, 1165–1171, 1178–1184, 1191–1200, 1207–1210
TOTAL	20602	3231	84%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
3960	225 💤	0 ❌	0 🔥	1m 10s ⏱️

[Copilot]

Pull request overview

Fixes Anthropic request option conflicts (notably allow_multiple_tool_calls and stream) and strengthens agent-based group chat manager parsing by adding bounded default retries and clearer retry guidance, with regression tests for the reported issues.

Changes:

• Filter unsupported/framework-level Anthropic options from prepared request payloads and keep stream explicit at SDK call sites.
• Default agent-orchestrator retry attempts to a bounded value and provide stricter retry instructions for non-JSON/invalid outputs.
• Add regression tests covering non-JSON manager output retry/raise behavior and Anthropic option filtering/stream conflicts.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
python/packages/anthropic/agent_framework_anthropic/_chat_client.py	Drops allow_multiple_tool_calls and stream from forwarded Anthropic request kwargs to prevent SDK TypeError conflicts.
python/packages/anthropic/tests/test_anthropic_client.py	Adds/updates tests ensuring stream is excluded from run options and doesn’t conflict with explicit streaming mode; validates allow_multiple_tool_calls handling.
python/packages/orchestrations/agent_framework_orchestrations/_group_chat.py	Sets a default retry budget for agent-based orchestration parsing and improves retry guidance (including participant name constraints).
python/packages/orchestrations/tests/test_group_chat.py	Adds tests for non-JSON manager output (retry then raise) and retry-then-recover behavior.