Fix an interrupted sub-agent not being properly resumed after the user answers "Yes" to `pattern_continue_interrupted`. This is fixed by introducing a transient `RESUMING` agent state.

Fixed `InvalidFlowIdException` and repeated error logs that occurred when a conversation tracker contained historical events referencing flows or slots that were subsequently removed from the model.

Updated security-flagged dependencies and allowed users to skip text normalization to reduce latency through the `no_text_normalization` parameter being passed correctly.

Fix bug when user's message during bot's speech is queued after bot is done speaking; improved handling of user interruptions based on minimum word settings; removed the 'hand over' command from the default GPT-5.1 prompt template.

Fixed bug where the validator was mutating the allowed values list of categorical slots, leading to corruption during training; improved handling of multi-agent flow interruptions.

Fixed tracing span attributes for LLM components using model groups; updated `cryptography` version to fix a security vulnerability.

Replaced generic server start message with IDE-specific instructions; added `rasa tools status` command displaying current project configuration.

Fixed correction reset inconsistency when a user responds to the active collect slot while correcting another slot.

Fixed three bugs in `ConcurrentRedisLockStore` causing crashes in cron jobs; adjusted Redis lock handling to prevent downstream caller crashes.

Breaking changes include updates to default language models and command generation prompts; new configurations for LLM model reasoning capabilities.