✦ Features 11
- support grouped content messagesbidirectional-streaming
- add strands update and polish compact, skills, rename, and setup flowscontextdevx
- add agent-level storage to BidiAgentbidirectional-streamingpersistence
- bring LocalAgent to parity with the TypeScript interfacedevxagent
- add text and reasoning stream eventsbidirectional-streaming
- support typing and speech in a shared consoledevxbidirectional-streaming
- preserve grouped tool calls through execution and deliverytoolbidirectional-streaming
- support cancellation from custom toolstoolbidirectional-streaming
- display tool calls in console I/Odevxbidirectional-streaming
- port a2a_client tool to TypeScripttoola2a
- graduate bidirectional streaming APIdevxbidirectional-streaming
⊘ Fixes 11
- include cache tokens in totalTokenscontextmodel
- map cache_write_tokens to cacheWriteInputTokensmodel
- defer OpenAI responses during user speechmodelbidirectional-streaming
- reject caching values other than 'auto', a bool, or Nonedevxmodel
- return portable filename from FileStorage.store instead of host pathpersistence
- deliver tool results that finish after a connection restarttoolbidirectional-streaming
- normalize tool inputs before replaymodeltool
- add context window limits for Sonnet 5.5, Opus 5.5, Fable 5.1 (py + ts)contextmodel
- skip location-source documents in OpenAIResponsesModel message formattingmodel
- keep stashed originals when backfilling after session resumecontextsessions
- remove context offloader plugin from harness in favor of context manager's stash
⋯ Other 7
- bump cedar-policy-mcp-schema-generator from 0.6.0 to 0.6.1 in /strands-py
- update default model
- remove session manager configurationbidirectional-streamingsessions
- add Human Overview sections to templates per AI usage reflection
- make tool execution internals privatedevxbidirectional-streaming
- use restart consistently for connection restartsbidirectional-streaming
- correct docstrings and drop unused barge-in and connection stop reasonsbidirectional-streaming