Shangjie Chen

This commit introduces a **new capability** to the **ADK** by establishing the `google.adk.environment` package, designed to manage code execution environments. It defines the abstract `BaseEnvironment` class and `ExecutionResult` dataclass, providing a standardized interface for operations like `execute`, `read_file`, and `write_file`. A concrete `LocalEnvironment` implementation is also included, enabling the ADK to perform local command execution and file system interactions using `asyncio` subprocesses. This foundational work creates a robust and extensible framework for managing diverse execution contexts within the ADK.

This commit performs a **configuration update** to the **`Event` Pydantic model** located in `src/google/adk/events/event.py`. It modifies the `model_config` to set `extra='ignore'`, which allows the model to **gracefully handle additional, undefined fields** during deserialization. This crucial change **prevents deserialization errors** when event data contains fields not explicitly defined in the model, significantly **improving compatibility**. Consequently, this enables the **session service to successfully accept and process 2.0 events**, ensuring robustness in event handling.

This commit introduces a **refactoring** within the **runner subsystem** (`src/google/adk/runners.py`) by implementing a new private factory method, `_create_invocation_context`. This factory is now responsible for creating `InvocationContext` instances, centralizing the object creation logic. The existing `_new_invocation_context` method has been updated to utilize this new factory, improving the internal code organization and maintainability of the **ADK runners**. This **chore** has no functional impact but enhances the consistency and extensibility of `InvocationContext` instantiation.

This commit **refactors the ADK runner** to automatically infer the `invocation_id` from `FunctionResponse` events when resuming a Human-in-the-Loop (HITL) flow. Specifically, the `run_async` method in `src/google/adk/runners.py` is enhanced to **resolve the invocation ID implicitly**, removing the need for it to be explicitly provided in most resuming scenarios. This **improves the usability and robustness** of the **ADK's flow execution and resumption logic**, particularly for scenarios involving tool confirmation. The change involves centralizing event matching helpers in `src/google/adk/flows/llm_flows/functions.py` and `src/google/adk/agents/invocation_context.py`, and is validated by a new test case for pausing and resuming tool confirmation without an explicit `invocation_id`.

This commit delivers a **bug fix** and **refactoring** to enhance **session management** within the `google.adk.sessions` module, specifically addressing stale session handling and potential deadlocks. It modifies the `database_session_service` to **aggressively reload stale sessions** from the database when the stored update time is newer, preventing transaction rejections. Crucially, synchronous SQLAlchemy introspection calls within `StorageSession` in `schemas/v0.py` and `schemas/v1.py` were **removed to prevent deadlocks** when used with `sqlalchemy.ext.asyncio` in Python 3.13. This change improves the overall stability and reliability of session operations by ensuring in-memory sessions are consistently synchronized with the database.

This commit **resolves a critical dependency compatibility issue** by explicitly adding `pypika>=0.50.0` to the project's `project.toml` file. This **dependency fix** is necessary because the `crewai` library, a key component, depends on `pypika`, and default older versions of `pypika` are incompatible with **Python 3.12 and newer environments**. By enforcing this minimum version, the change **ensures stable installation and correct runtime behavior** for the project across modern Python versions.

This commit introduces a **new capability** to **filter out specific `adk_request_input` events** from the content list used in **LLM flow context generation**. It modifies the `google.adk.flows.llm_flows.contents` module by adding `_is_request_input_event` to exclude these events via `_should_include_event_in_context`. A new constant, `REQUEST_INPUT_FUNCTION_CALL_NAME`, is also defined in `google.adk.flows.llm_flows.functions` to consistently identify these events. This ensures that `adk_request_input` events are no longer included in the context provided to LLM models, thereby **refining the input data** and potentially improving model efficiency.

This commit introduces a **new capability** to refine the **agent context management** within the **ADK (Agent Development Kit) LLM flows**. It adds a new internal function, `_is_adk_framework_event`, to `src/google/adk/flows/llm_flows/contents.py` to specifically identify events originating from an internal `adk_framework` function. This mechanism is integrated into the `_should_include_event_in_context` logic, ensuring these internal framework responses are **excluded from the agent's operational context**. The change prevents internal framework details from polluting the agent's understanding, thereby improving the **relevance and focus of the agent's decision-making**.

This commit **rolls back a breaking database migration** that was causing instability within the **database session service**. It specifically removes schema migration checks, deletes the `StorageMetadata` table, and restructures the `StorageEvent` in `src/google/adk/sessions/database_session_service.py` to store event data in individual columns rather than a single JSON field. This **maintenance fix** ensures the database schema is stable and the session service operates correctly. Additionally, new unit tests were introduced for `DynamicPickleType` to verify its behavior across different database dialects, enhancing the robustness of data handling.

This commit **adds a new migration guide** for the **DatabaseSessionService**, specifically detailing the process for introducing new schema versions. A new file, `src/google/adk/sessions/migration/README.md`, is introduced to outline these steps. This **documentation addition** is a **chore** that aims to standardize and clarify the procedure for evolving the database session schema. The guide will help developers ensure proper handling of schema changes, improving the maintainability and future compatibility of the `DatabaseSessionService`.

This commit introduces a **major database schema migration** for the **session management subsystem**, refactoring the `DatabaseSessionService` to utilize a **JSON serialization schema** for storing session events, replacing the previous SQLAlchemy Pickle format. This **new capability** enhances data portability and future-proofs session storage by introducing a `StorageMetadata` table for **schema versioning**. To facilitate this transition, a **new command-line tool**, `adk migrate session`, has been added, enabling users to seamlessly **migrate existing databases** from the old schema to the new JSON-based format. This change impacts all users of `DatabaseSessionService`, requiring a migration step for existing deployments to ensure continued compatibility and leverage the improved session storage.

This commit introduces a **new feature** to the **agents subsystem**, specifically enhancing the `BaseAgent` class. A new field validator, `validate_sub_agents_unique_names`, has been added to detect and log a warning when **duplicate sub-agent names** are encountered during agent configuration. This **validation enhancement** improves the robustness of agent definitions by providing early feedback on potential naming conflicts, helping users prevent unexpected behavior. Comprehensive unit tests were also added to ensure the validator's correctness across various scenarios.

This commit delivers a **bug fix** by **removing a destructive validation step** from the system. This particular validation was causing unintended issues, potentially leading to data corruption or operational failures. By eliminating this problematic check, the change aims to **improve the overall stability and reliability** of the affected module's operations, preventing future negative impacts on data or user workflows.

chore: Update MCP requirement to >1.10.0

chore: Allow google-cloud-storage >=2.18.0

This commit performs a **routine maintenance chore** by **bumping the project version** for the `google.adk` library from `v1.18.0` to `v1.19.0`. Specifically, it updates the `__version__` string within `src/google/adk/version.py` to reflect the new release number. This change primarily affects the **core versioning mechanism** of the library, signaling the start of a new development cycle and preparing for the release of new features or fixes, as typically documented in the `CHANGELOG.md`.

This commit introduces a **new BigQuery-related label**, 'bq', into the project's automated triaging system. It updates the **`adk_triaging_agent`** within `contributing/samples/adk_triaging_agent/agent.py` to incorporate this new label. This **maintenance chore** involves modifying the agent's owner mapping, guidelines, and labeling instructions to recognize and apply the 'bq' label. The change aims to improve the automated classification and routing of **BigQuery-related issues**, streamlining the triaging process for relevant teams.

This commit **adds a new Python sample script** to demonstrate the **rewind session functionality** of the agent. The new sample, located at `contributing/samples/rewind_session/main.py`, serves as a practical example for developers. This **new feature** provides a clear, runnable illustration of how to utilize the rewind session, including calls to `highlight` and `call_agent_async`. It enhances the project's documentation and usability by offering a concrete example for understanding and integrating this core agent capability.

This commit provides a **bug fix** for the **`vertex_ai_session_service`** by modifying the `get_session` method to use `async for` when iterating through the event stream. This change ensures that all events from the asynchronous iterator are properly processed and retrieved, addressing an issue where sessions with numerous events might not have been fully loaded. The update **ensures reliable retrieval of all events** for a session, preventing incomplete data. New unit tests were added to `test_vertex_ai_session_service.py` to specifically validate the correct handling of sessions with a large number of events.

This commit introduces a **new capability** to the **ADK framework**, enabling users to register **custom backend services** for sessions, memory, and artifacts without modifying the core ADK. This **enhances extensibility** by allowing services to be defined in local `services.yaml` or `services.py` files, which are automatically discovered and loaded. The `src/google/adk/cli/service_registry.py` module now orchestrates this loading and registration process, integrating these custom services into the ADK's FastAPI application via `src/google/adk/cli/fast_api.py`. This provides a flexible mechanism for **custom backend integration**, allowing users to tailor ADK deployments with their specific service implementations.